idnits 2.17.00 (12 Aug 2021) /tmp/idnits46868/draft-ietf-tewg-principles-00.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- ** Looks like you're using RFC 2026 boilerplate. This must be updated to follow RFC 3978/3979, as updated by RFC 4748. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- ** The document seems to lack a 1id_guidelines paragraph about the list of current Internet-Drafts. ** The document seems to lack a 1id_guidelines paragraph about the list of Shadow Directories. == No 'Intended status' indicated for this document; assuming Proposed Standard == The page length should not exceed 58 lines per page, but there was 60 longer pages, the longest (page 1) being 62 lines Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- ** The document seems to lack an IANA Considerations section. (See Section 2.2 of https://www.ietf.org/id-info/checklist for how to handle the case when there are no actions for IANA.) ** The document seems to lack separate sections for Informative/Normative References. All references will be assumed normative when checking for downward references. Miscellaneous warnings: ---------------------------------------------------------------------------- == Line 670 has weird spacing: '...ay) and timin...' == Line 777 has weird spacing: '...ormance optim...' == Line 888 has weird spacing: '... merits of di...' == Line 1542 has weird spacing: '...provide bound...' -- The document seems to lack a disclaimer for pre-RFC5378 work, but may have content which was first submitted before 10 November 2008. If you have contacted all the original authors and they are all willing to grant the BCP78 rights to the IETF Trust, then this is fine, and you can ignore this comment. If not, you may need to add the pre-RFC5378 disclaimer. (See the Legal Provisions document at https://trustee.ietf.org/license-info for more information.) -- Couldn't find a document date in the document -- date freshness check skipped. Checking references for intended status: Proposed Standard ---------------------------------------------------------------------------- (See RFCs 3967 and 4897 for information about using normative references to lower-maturity documents in RFCs) == Missing Reference: 'Elwalid' is mentioned on line 312, but not defined == Missing Reference: 'RFC 2475' is mentioned on line 656, but not defined == Missing Reference: 'RFC-2474' is mentioned on line 1633, but not defined == Missing Reference: 'RFC2330' is mentioned on line 1720, but not defined == Missing Reference: 'RFC2680' is mentioned on line 1721, but not defined ** Obsolete undefined reference: RFC 2680 (Obsoleted by RFC 7680) == Missing Reference: 'RFC2679' is mentioned on line 1721, but not defined ** Obsolete undefined reference: RFC 2679 (Obsoleted by RFC 7679) == Missing Reference: 'RFC2678' is mentioned on line 1722, but not defined == Missing Reference: 'BGP4' is mentioned on line 2677, but not defined == Unused Reference: 'ASH1' is defined on line 2853, but no explicit reference was found in the text == Unused Reference: 'AWD4' is defined on line 2873, but no explicit reference was found in the text == Unused Reference: 'AWD5' is defined on line 2876, but no explicit reference was found in the text == Unused Reference: 'ELW95' is defined on line 2887, but no explicit reference was found in the text == Unused Reference: 'FGLR' is defined on line 2892, but no explicit reference was found in the text == Unused Reference: 'LNO96' is defined on line 2922, but no explicit reference was found in the text == Unused Reference: 'RFC-1458' is defined on line 2946, but no explicit reference was found in the text == Unused Reference: 'RFC-1771' is defined on line 2949, but no explicit reference was found in the text == Unused Reference: 'RFC-1812' is defined on line 2952, but no explicit reference was found in the text == Unused Reference: 'RFC-1997' is defined on line 2958, but no explicit reference was found in the text == Unused Reference: 'RFC-1998' is defined on line 2961, but no explicit reference was found in the text == Unused Reference: 'RFC-2215' is defined on line 2975, but no explicit reference was found in the text == Unused Reference: 'RFC-2216' is defined on line 2979, but no explicit reference was found in the text == Unused Reference: 'RFC-2330' is defined on line 2982, but no explicit reference was found in the text == Unused Reference: 'RFC-2678' is defined on line 2995, but no explicit reference was found in the text == Unused Reference: 'RFC-2679' is defined on line 2998, but no explicit reference was found in the text == Unused Reference: 'RFC-2680' is defined on line 3001, but no explicit reference was found in the text -- Possible downref: Non-RFC (?) normative reference: ref. 'ASH1' -- Possible downref: Non-RFC (?) normative reference: ref. 'ASH2' -- Possible downref: Non-RFC (?) normative reference: ref. 'ASH3' -- Possible downref: Non-RFC (?) normative reference: ref. 'AWD1' -- Possible downref: Non-RFC (?) normative reference: ref. 'AWD2' -- Possible downref: Non-RFC (?) normative reference: ref. 'AWD3' -- Possible downref: Non-RFC (?) normative reference: ref. 'AWD4' -- Possible downref: Non-RFC (?) normative reference: ref. 'AWD5' -- Possible downref: Non-RFC (?) normative reference: ref. 'CRUZ' -- Possible downref: Non-RFC (?) normative reference: ref. 'DIFF-TE' -- Possible downref: Non-RFC (?) normative reference: ref. 'ELW95' -- Possible downref: Non-RFC (?) normative reference: ref. 'FGLR' -- Possible downref: Non-RFC (?) normative reference: ref. 'FLJA93' -- Possible downref: Non-RFC (?) normative reference: ref. 'FLOY94' -- Possible downref: Non-RFC (?) normative reference: ref. 'HUSS87' -- Possible downref: Non-RFC (?) normative reference: ref. 'ITU-E600' -- Possible downref: Non-RFC (?) normative reference: ref. 'ITU-E701' -- Possible downref: Non-RFC (?) normative reference: ref. 'ITU-E801' -- Possible downref: Non-RFC (?) normative reference: ref. 'JAM' -- Possible downref: Non-RFC (?) normative reference: ref. 'KATZ' -- Possible downref: Non-RFC (?) normative reference: ref. 'LNO96' -- Possible downref: Non-RFC (?) normative reference: ref. 'MA' -- Possible downref: Non-RFC (?) normative reference: ref. 'MATE' -- Possible downref: Non-RFC (?) normative reference: ref. 'MCQ80' -- Possible downref: Non-RFC (?) normative reference: ref. 'MPLS-DIFF' -- Possible downref: Non-RFC (?) normative reference: ref. 'MR99' ** Obsolete normative reference: RFC 1349 (Obsoleted by RFC 2474) ** Downref: Normative reference to an Informational RFC: RFC 1458 ** Obsolete normative reference: RFC 1771 (Obsoleted by RFC 4271) ** Downref: Normative reference to an Informational RFC: RFC 1992 ** Downref: Normative reference to an Informational RFC: RFC 1998 ** Obsolete normative reference: RFC 2178 (Obsoleted by RFC 2328) ** Downref: Normative reference to an Informational RFC: RFC 2216 ** Downref: Normative reference to an Informational RFC: RFC 2330 ** Downref: Normative reference to an Informational RFC: RFC 2386 ** Downref: Normative reference to an Informational RFC: RFC 2475 ** Obsolete normative reference: RFC 2679 (Obsoleted by RFC 7679) ** Obsolete normative reference: RFC 2680 (Obsoleted by RFC 7680) ** Downref: Normative reference to an Informational RFC: RFC 2702 ** Downref: Normative reference to an Informational RFC: RFC 2722 ** Downref: Normative reference to an Informational RFC: RFC 2753 ** Downref: Normative reference to an Informational RFC: RFC 2998 ** Downref: Normative reference to an Informational RFC: RFC 3086 -- Possible downref: Non-RFC (?) normative reference: ref. 'SHAR' -- Possible downref: Non-RFC (?) normative reference: ref. 'SLDC98' -- Possible downref: Non-RFC (?) normative reference: ref. 'SMIT' -- Possible downref: Non-RFC (?) normative reference: ref. 'XIAO' -- Possible downref: Non-RFC (?) normative reference: ref. 'YARE95' Summary: 24 errors (**), 0 flaws (~~), 31 warnings (==), 33 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Internet Engineering Task Force 3 INTERNET-DRAFT 4 TE Working Group 5 Daniel O. Awduche 6 Expiration Date: February 2002 Movaz Networks 8 Angela Chiu 9 Celion Networks 11 Anwar Elwalid 12 Lucent Technologies 14 Indra Widjaja 15 Lucent Technologies 17 XiPeng Xiao 18 Photuris 20 Overview and Principles of Internet Traffic Engineering 22 draft-ietf-tewg-principles-00.txt 24 Status of this Memo 26 This document is an Internet-Draft and is in full conformance with 27 all provisions of Section 10 of RFC2026. 29 Internet-Drafts are working documents of the Internet Engineering 30 Task Force (IETF), its areas, and its working groups. Note that 31 other groups may also distribute working documents as Internet- 32 Drafts. 34 Internet-Drafts are draft documents valid for a maximum of six months 35 and may be updated, replaced, or obsoleted by other documents at any 36 time. It is inappropriate to use Internet-Drafts as reference 37 material or to cite them other than as "work in progress." 39 To view the list Internet-Draft Shadow Directories, see 40 http://www.ietf.org/shadow.html. 42 Abstract 44 This memo describes principles for Traffic Engineering (TE) in the 45 Internet. The document is intended to promote better understanding 46 of the issues surrounding traffic engineering in IP networks, and to 47 provide a common basis for the development of traffic engineering 48 capabilities for the Internet. The principles, architectures, and 49 methodologies for performance evaluation and performance optimization 50 of operational IP networks are discussed throughout this document. 51 The optimization goals of traffic engineering are to enhance the 52 performance of IP traffic while utilizing network resources 53 economically and reliably. The document includes a set of generic 54 recommendations, and options for Internet traffic engineering. The 55 document can serve as a guide to implementors of online and offline 56 Internet traffic engineering mechanisms, tools, and support systems. 57 The document can also help service providers devise traffic 58 engineering solutions for their networks. 60 Table of Contents 62 1.0 Introduction...................................................3 63 1.1 What is Internet Traffic Engineering?.......................4 64 1.2 Scope.......................................................7 65 1.3 Terminology.................................................8 66 2.0 Background....................................................11 67 2.1 Context of Internet Traffic Engineering....................11 68 2.2 Network Context............................................12 69 2.3 Problem Context............................................14 70 2.3.1 Congestion and its Ramifications......................15 71 2.4 Solution Context...........................................15 72 2.4.1 Combating the Congestion Problem......................17 73 2.5 Implementation and Operational Context.....................19 74 3.0 Traffic Engineering Process Model.............................20 75 3.1 Components of the Traffic Engineering Process Model........21 76 3.2 Measurement................................................21 77 3.3 Modeling, Analysis, and Simulation.........................22 78 3.4 Optimization...............................................23 79 4.0 Historical Review and Recent Developments.....................24 80 4.1 Traffic Engineering in Classical Telephone Networks........24 81 4.2 Evolution of Traffic Engineering in the Internet...........26 82 4.2.1 Adaptive Routing in ARPANET...........................26 83 4.2.2 Dynamic Routing in the Internet.......................27 84 4.2.3 ToS Routing...........................................27 85 4.2.4 Equal Cost Multi-Path.................................28 86 4.2.5 Nimrod................................................28 87 4.3 Overlay Model..............................................29 88 4.4 Constraint-Based Routing...................................29 89 4.5 Overview of Other IETF Projects Related to Traffic 90 Engineering................................................30 91 4.5.1 Integrated Services...................................30 92 4.5.2 RSVP..................................................31 93 4.5.3 Differentiated Services...............................32 94 4.5.4 MPLS..................................................33 95 4.5.5 IP Performance Metrics................................34 96 4.5.6 Flow Measurement......................................34 97 4.5.7 Endpoint Congestion Management........................35 98 4.6 Overview of ITU Activities Related to Traffic 99 Engineering................................................35 100 4.7 Content Distribution.......................................36 101 5.0 Taxonomy of Traffic Engineering Systems.......................37 102 5.1 Time-Dependent Versus State-Dependent......................37 103 5.2 Offline Versus Online......................................38 104 5.3 Centralized Versus Distributed.............................38 105 5.4 Local Versus Global........................................39 106 5.5 Prescriptive Versus Descriptive............................39 107 5.6 Open-Loop Versus Closed-Loop...............................40 108 5.7 Tactical vs Strategic......................................40 109 6.0 Recommendations for Internet Traffic Engineering..............40 110 6.1 Generic Non-functional Recommendations.....................41 111 6.2 Routing Recommendations....................................42 112 6.3 Traffic Mapping Recommendations............................45 113 6.4 Measurement Recommendations................................45 114 6.5 Network Survivability......................................46 115 6.5.1 Survivability in MPLS Based Networks..................48 116 6.5.2 Protection Option.....................................49 117 6.6 Traffic Engineering in Diffserv Environments...............50 118 6.7 Network Controllability....................................52 119 7.0 Inter-Domain Considerations...................................52 120 8.0 Overview of Contemporary TE Practices in Operational 121 IP Networks...................................................53 122 9.0 Conclusion....................................................55 123 10.0 Security Considerations......................................55 124 11.0 Acknowledgments..............................................55 125 12.0 References...................................................56 126 13.0 Authors' Addresses...........................................60 128 1.0 Introduction 130 This memo describes principles for Internet traffic engineering. The 131 objective of the document is to articulate the general issues and 132 principles for Internet traffic engineering; and where appropriate to 133 provide recommendations, guidelines, and options for the development 134 of online and offline Internet traffic engineering capabilities and 135 support systems. 137 The document can aid service providers in devising and implementing 138 traffic engineering solutions for their networks. Networking hardware 139 and software vendors will also find the document helpful in the 140 development of mechanisms and support systems for the Internet 141 environment that support the traffic engineering function. 143 The document provides a terminology for describing and understanding 144 common Internet traffic engineering concepts. The document also 145 provides a taxonomy of known traffic engineering styles. In this 146 context, a traffic engineering style abstracts important aspects from 147 a traffic engineering methodology. Traffic engineering styles can be 148 viewed in different ways depending upon the specific context in which 149 they are used and the specific purpose which they serve. The 150 combination of styles and views results in a natural taxonomy of 151 traffic engineering systems. 153 Even though Internet traffic engineering is most effective when 154 applied end-to-end, the initial focus of this document document is 155 intra-domain traffic engineering (that is, traffic engineering within 156 a given autonomous system). However, because a preponderance of 157 Internet traffic tends to be inter-domain (originating in one 158 autonomous system and terminating in another), this document provides 159 an overview of aspects pertaining to inter-domain traffic 160 engineering. 162 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 163 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 164 document are to be interpreted as described in RFC 2119. 166 1.1. What is Internet Traffic Engineering? 168 Internet traffic engineering is defined as that aspect of Internet 169 network engineering dealing with the issue of performance evaluation 170 and performance optimization of operational IP networks. Traffic 171 Engineering encompasses the application of technology and scientific 172 principles to the measurement, characterization, modeling, and 173 control of Internet traffic [RFC-2702, AWD2]. 175 Enhancing the performance of an operational network, at both the 176 traffic and resource levels, are major objectives of Internet traffic 177 engineering. This is accomplished by addressing traffic oriented 178 performance requirements, while utilizing network resources 179 economically and reliably. Traffic oriented performance measures 180 include delay, delay variation, packet loss, and throughput. 182 An important objective of Internet traffic engineering is to 183 facilitate reliable network operations [RFC-2702]. Reliable network 184 operations can be facilitated by providing mechanisms that enhance 185 network integrity and by embracing policies emphasizing network 186 survivability. This results in a minimization of the vulnerability of 187 the network to service outages arising from errors, faults, and 188 failures occurring within the infrastructure. 190 The Internet exists in order to transfer information from source 191 nodes to destination nodes. Accordingly, one of the most significant 192 functions performed by the Internet is the routing of traffic from 193 ingress nodes to egress nodes. Therefore, one of the most distinctive 194 functions performed by Internet traffic engineering is the control 195 and optimization of the routing function, to steer traffic through 196 the network in the most effective way. 198 Ultimately, it is the performance of the network as seen by end users 199 of network services that is truly paramount. This crucial point 200 should be considered throughout the development of traffic 201 engineering mechanisms and policies. The characteristics visible to 202 end users are the emergent properties of the network, which are the 203 characteristics of the network when viewed as a whole. A central goal 204 of the service provider, therefore, is to enhance the emergent 205 properties of the network while taking economic considerations into 206 account. 208 The importance of the above observation regarding the emergent 209 properties of networks is that special care must be taken when 210 choosing network performance measures to optimize. Optimizing the 211 wrong measures may achieve certain local objectives, but may have 212 disastrous consequences on the emergent properties of the network and 213 thereby on the quality of service perceived by end-users of network 214 services. 216 A subtle, but practical advantage of the systematic application of 217 traffic engineering concepts to operational networks is that it helps 218 to identify and structure goals and priorities in terms of enhancing 219 the quality of service delivered to end-users of network services. 220 The application of traffic engineering concepts also aids in the 221 measurement and analysis of the achievement of these goals. 223 The optimization aspects of traffic engineering can be achieved 224 through capacity management and traffic management. As used in this 225 document, capacity management includes capacity planning, routing 226 control, and resource management. Network resources of particular 227 interest include link bandwidth, buffer space, and computational 228 resources. Likewise, as used in this document, traffic management 229 includes (1) nodal traffic control functions such as traffic 230 conditioning, queue management, scheduling, and (2) other functions 231 that regulate traffic flow through the network or that arbitrate 232 access to network resources between different packets or between 233 different traffic streams. 235 The optimization objectives of Internet traffic engineering should be 236 viewed as a continual and iterative process of network performance 237 improvement and not simply as a one time goal. Traffic engineering 238 also demands continual development of new technologies and new 239 methodologies for network performance enhancement. 241 The optimization objectives of Internet traffic engineering may 242 change over time as new requirements are imposed, as new technologies 243 emerge, or as new insights are brought to bear on the underlying 244 problems. Moreover, different networks may have different 245 optimization objectives, depending upon their business models, 246 capabilities, and operating constraints. The optimization aspects of 247 traffic engineering are ultimately concerned with network control 248 regardless of the specific optimization goals in any particular 249 environment. 251 Thus, the optimization aspects of traffic engineering can be viewed 252 from a control perspective. The aspect of control within the Internet 253 traffic engineering arena can be pro-active and/or reactive. In the 254 pro-active case, the traffic engineering control system takes 255 preventive action to obviate predicted unfavorable future network 256 states. It may also take perfective action to induce a more 257 desirable state in the future. In the reactive case, the control 258 system responds correctively and perhaps adaptively to events that 259 have already transpired in the network. 261 The control dimension of Internet traffic engineering responds at 262 multiple levels of temporal resolution to network events. Certain 263 aspects of capacity management, such as capacity planning, respond at 264 very coarse temporal levels, ranging from days to possibly years. The 265 introduction of automatically switched optical transport networks 266 (e.g. based on the Multi-protocol Lambda Switching concepts) could 267 significantly reduce the lifecycle for capacity planning by 268 expediting provisioning of optical bandwidth. Routing control 269 functions operate at intermediate levels of temporal resolution, 270 ranging from milliseconds to days. Finally, the packet level 271 processing functions (e.g. rate shaping, queue management, and 272 scheduling) operate at very fine levels of temporal resolution, 273 ranging from picoseconds to milliseconds while responding to the 274 real-time statistical behavior of traffic. The subsystems of Internet 275 traffic engineering control include: capacity augmentation, routing 276 control, traffic control, and resource control (including control of 277 service policies at network elements). When capacity is to be 278 augmented for tactical purposes, it may be desirable to devise a 279 deployment plan that expedites bandwidth provisioning while 280 minimizing installation costs. 282 Inputs into the traffic engineering control system include network 283 state variables, policy variables, and decision variables. 285 One major challenge of Internet traffic engineering is the 286 realization of automated control capabilities that adapt quickly and 287 cost effectively to significant changes in a network's state, while 288 still maintaining stability. 290 Another critical dimension of Internet traffic engineering is network 291 performance evaluation, which is important for assessing the 292 effectiveness of traffic engineering methods, and for monitoring and 293 verifying compliance with network performance goals. Results from 294 performance evaluation can be used to identify existing problems, 295 guide network re-optimization, and aid in the prediction of potential 296 future problems. 298 Performance evaluation can be achieved in many different ways. The 299 most notable techniques include analytical methods, simulation, and 300 empirical methods based on measurements. When analytical methods or 301 simulation are used, network nodes and links can be modeled to 302 capture relevant operational features such as topology, bandwidth, 303 buffer space, and nodal service policies (link scheduling, packet 304 prioritization, buffer management, etc). Analytical traffic models 305 can be used to depict dynamic and behavioral traffic characteristics, 306 such as burstiness, statistical distributions, and dependence. 308 Performance evaluation can be quite complicated in practical network 309 contexts. A number of techniques can be used to simplify the 310 analysis, such as abstraction, decomposition, and approximation. For 311 example, simplifying concepts such as effective bandwidth and 312 effective buffer [Elwalid] may be used to approximate nodal behaviors 313 at the packet level and simplify the analysis at the connection 314 level. Network analysis techniques using, for example, queuing models 315 and approximation schemes based on asymptotic and decomposition 316 techniques can render the analysis even more tractable. In 317 particular, an emerging set of concepts known as network calculus 318 [CRUZ] based on deterministic bounds may simplify network analysis 319 relative to classical stochastic techniques. When using analytical 320 techniques, care should be taken to ensure that the models faithfully 321 reflect the relevant operational characteristics of the modeled 322 network entities. 324 Simulation can be used to evaluate network performance or to verify 325 and validate analytical approximations. Simulation can, however, be 326 computationally costly and may not always provide sufficient 327 insights. An appropriate approach to a given network performance 328 evaluation problem may involve a hybrid combination of analytical 329 techniques, simulation, and empirical methods. 331 As a general rule, traffic engineering concepts and mechanisms must 332 be sufficiently specific and well defined to address known 333 requirements, but simultaneously flexible and extensible to 334 accommodate unforeseen future demands. 336 1.2. Scope 338 The scope of this document is intra-domain traffic engineering; that 339 is, traffic engineering within a given autonomous system in the 340 Internet. The document will discuss concepts pertaining to intra- 341 domain traffic control, including such issues as routing control, 342 micro and macro resource allocation, and the control coordination 343 problems that arise consequently. 345 This document will describe and characterize techniques already in 346 use or in advanced development for Internet traffic engineering. The 347 way these techniques fit together will be discussed and scenarios in 348 which they are useful will be identified. 350 Although the emphasis is on intra-domain traffic engineering, in 351 Section 7.0, an overview of the high level considerations pertaining 352 to inter-domain traffic engineering will be provided. inter-domain 353 Internet traffic engineering is crucial to the performance 354 enhancement of the global Internet infrastructure. 356 Whenever possible, relevant requirements from existing IETF documents 357 and other sources will be incorporated by reference. 359 1.3 Terminology 361 This subsection provides terminology which is useful for Internet 362 traffic engineering. The definitions presented apply to this 363 document. These terms may have other meanings elsewhere. 365 - Baseline analysis: 366 A study conducted to serve as a baseline for comparison to the 367 actual behavior of the network. 369 - Busy hour: 370 A one hour period within a specified interval of time 371 (typically 24 hours) in which the traffic load in a 372 network or sub-network is greatest. 374 - Bottleneck 375 A network element whose input traffic rate tends to be greater 376 than its output rate. 378 - Congestion: 379 A state of a network resource in which the traffic incident 380 on the resource exceeds its output capacity over an interval 381 of time. 383 - Congestion avoidance: 384 An approach to congestion management that attempts to obviate 385 the occurrence of congestion. 387 - Congestion control: 388 An approach to congestion management that attempts to remedy 389 congestion problems that have already occurred. 391 - Constraint-based routing: 392 A class of routing protocols that take specified traffic 393 attributes, network constraints, and policy constraints into 394 account when making routing decisions. Constraint-based 395 routing is applicable to traffic aggregates as well as flows. 396 It is a generalization of QoS routing. 398 - Demand side congestion management: 399 A congestion management scheme that addresses congestion 400 problems by regulating or conditioning offered load. 402 - Effective bandwidth: 403 The minimum amount of bandwidth that can be assigned to a flow 404 or traffic aggregate in order to deliver 'acceptable service 405 quality' to the flow or traffic aggregate. 407 - Egress traffic: 408 Traffic exiting a network or network element. 410 - Hot-spot 411 A network element or subsystem which is in a state of 412 congestion. 414 - Ingress traffic: 415 Traffic entering a network or network element. 417 - Inter-domain traffic: 418 Traffic that originates in one Autonomous system and 419 terminates in another. 421 - Loss network: 422 A network that does not provide adequate buffering for 423 traffic, so that traffic entering a busy resource within 424 the network will be dropped rather than queued. 426 - Metric: 427 A parameter defined in terms of standard units of 428 measurement. 430 - Measurement Methodology: 431 A repeatable measurement technique used to derive one or 432 more metrics of interest. 434 - Network Survivability: 435 The capability to provide a prescribed level of QoS for 436 existing services after a given number of failures occur 437 within the network. 439 - Offline traffic engineering: 440 A traffic engineering system that exists outside of the 441 network. 443 - Online traffic engineering: 444 A traffic engineering system that exists within the network, 445 typically implemented on or as adjuncts to operational network 446 elements. 448 - Performance measures: 449 Metrics that provide quantitative or qualitative measures of 450 the performance of systems or subsystems of interest. 452 - Performance management: 453 A systematic approach to improving effectiveness in the 454 accomplishment of specific networking goals related to 455 performance improvement. 457 - Performance Metric: 458 A performance parameter defined in terms of standard units of 459 measurement. 461 - Provisioning: 462 The process of assigning or configuring network resources to 463 meet certain requests. 465 - QoS routing: 467 Class of routing systems that selects paths to be used by a 468 flow based on the QoS requirements of the flow. 470 - Service Level Agreement: 471 A contract between a provider and a customer that guarantees 472 specific levels of performance and reliability at a certain 473 cost. 475 - Stability: 476 An operational state in which a network does not oscillate 477 in a disruptive manner from one mode to another mode. 479 - Supply side congestion management: 480 A congestion management scheme that provisions additional 481 network resources to address existing and/or anticipated 482 congestion problems. 484 - Transit traffic: 485 Traffic whose origin and destination are both outside of 486 the network under consideration. 488 - Traffic characteristic: 489 A description of the temporal behavior or a description of the 490 attributes of a given traffic flow or traffic aggregate. 492 - Traffic engineering system 493 A collection of objects, mechanisms, and protocols that are 494 used conjunctively to accomplish traffic engineering 495 objectives. 497 - Traffic flow: 498 A stream of packets between two end-points that can be 499 characterized in a certain way. A micro-flow has a more 500 specific definition: A micro-flow is a stream of packets 501 with the same source and destination addresses, source 502 and destination ports, and protocol ID. 504 - Traffic intensity: 505 A measure of traffic loading with respect to a resource 506 capacity over a specified period of time. In classical 507 telephony systems, traffic intensity is measured in units of 508 Erlang. 510 - Traffic matrix: 511 A representation of the traffic demand between a set of origin 512 and destination abstract nodes. An abstract node can consist 513 of one or more network elements. 515 - Traffic monitoring: 517 The process of observing traffic characteristics at a given 518 point in a network and collecting the traffic information for 519 analysis and further action. 521 - Traffic trunk: 522 An aggregation of traffic flows belonging to the same class 523 which are forwarded through a common path. A traffic trunk 524 may be characterized by an ingress and egress node, and a 525 set of attributes which determine its behavioral 526 characteristics and requirements from the network. 528 2.0 Background 530 The Internet has quickly evolved into a very critical communications 531 infrastructure, supporting significant economic, educational, and 532 social activities. Simultaneously, the delivery of Internet 533 communications services has become very competitive and end-users are 534 demanding very high quality service from their service providers. 535 Consequently, performance optimization of large scale IP networks, 536 especially public Internet backbones, has become an important 537 problem. Network performance requirements are multi-dimensional, 538 complex, and sometimes contradictory; making the traffic engineering 539 problem very challenging. 541 The network must convey IP packets from ingress nodes to egress nodes 542 efficiently, expeditiously and economically. Furthermore, in a 543 multiclass service environment (e.g. Diffserv capable networks), the 544 resource sharing parameters of the network must be appropriately 545 determined and configured according to prevailing policies and 546 service models to resolve resource contention issues arising from 547 mutual interference between packets traversing through the network. 548 Thus, consideration must be given to resolving competition for 549 network resources between traffic streams belonging to the same 550 service class (intra-class contention resolution) and traffic streams 551 belonging to different classes (inter-class contention resolution). 553 2.1 Context of Internet Traffic Engineering 555 The context of Internet traffic engineering pertains to the scenarios 556 where traffic engineering is used. A traffic engineering methodology 557 establishes appropriate rules to resolve traffic performance issues 558 occurring in a specific context. The context of Internet traffic 559 engineering includes: 561 (1) A network context defining the universe of discourse, 562 and in particular the situations in which the traffic 563 engineering problems occur. The network context 564 includes network structure, network policies, network 565 characteristics, network constraints, network quality 566 attributes, and network optimization criteria. 568 (2) A problem context defining the general and concrete 569 issues that traffic engineering addresses. The problem 570 context includes identification, abstraction of relevant 571 features, representation, formulation, specification of 572 the requirements on the solution space, and specification 573 of the desirable features of acceptable solutions. 575 (3) A solution context suggesting how to address the issues 576 identified by the problem context. The solution context 577 includes analysis, evaluation of alternatives, 578 prescription, and resolution. 580 (4) An implementation and operational context in which the 581 solutions are methodologically instantiated. The 582 implementation and operational context includes 583 planning, organization, and execution. 585 The context of Internet traffic engineering and the different problem 586 scenarios are discussed in the following subsections. 588 2.2 Network Context 590 IP networks range in size from small clusters of routers situated 591 within a given location, to thousands of interconnected routers, 592 switches, and other components distributed all over the world. 594 Conceptually, at the most basic level of abstraction, an IP network 595 can be represented as a distributed dynamical system consisting of: 596 (1) a set of interconnected resources which provide transport 597 services for IP traffic subject to certain constraints, (2) a demand 598 system representing the offered load to be transported through the 599 network, and (3) a response system consisting of network processes, 600 protocols, and related mechanisms which facilitate the movement of 601 traffic through the network [see also AWD2]. 603 The network elements and resources may have specific characteristics 604 restricting the manner in which the demand is handled. Additionally, 605 network resources may be equipped with traffic control mechanisms 606 superintending the way in which the demand is serviced. Traffic 607 control mechanisms may, for example, be used to control various 608 packet processing activities within a given resource, arbitrate 609 contention for access to the resource by different packets, and 610 regulate traffic behavior through the resource. A configuration 611 management and provisioning system may allow the settings of the 612 traffic control mechanisms to be manipulated by external or internal 613 entities in order to exercise control over the way in which the 614 network elements respond to internal and external stimuli. 616 The details of how the network provides transport services for 617 packets are specified in the policies of the network administrators 618 and are installed through network configuration management and policy 619 based provisioning systems. Generally, the types of services 620 provided by the network also depends upon the technology and 621 characteristics of the network elements and protocols, the prevailing 622 service and utility models, and the ability of the network 623 administrators to translate policies into network configurations. 625 Contemporary Internet networks have three significant 626 characteristics: (1) they provide real-time services, (2) they have 627 become mission critical, and (3) their operating environments are 628 very dynamic. The dynamic characteristics of IP networks can be 629 attributed in part to fluctuations in demand, to the interaction 630 between various network protocols and processes, to the rapid 631 evolution of the infrastructure which demands the constant inclusion 632 of new technologies and new network elements, and to transient and 633 persistent impairments which occur within the system. 635 Packets contend for the use of network resources as they are conveyed 636 through the network. A network resource is considered to be 637 congested if the arrival rate of packets exceed the output capacity 638 of the resource over an interval of time. Congestion may result in 639 some of the arrival packets being delayed or even dropped. 640 Congestion increases transit delays, delay variation, packet loss, 641 and reduces the predictability of network services. Clearly, 642 congestion is a highly undesirable phenomenon. 644 Combating congestion at reasonable cost is a major objective of 645 Internet traffic engineering. 647 Efficient sharing of network resources by multiple traffic streams is 648 a basic economic premise for packet switched networks in general and 649 the Internet in particular. A fundamental challenge in network 650 operation, especially in a large scale public IP network, is to 651 increase the efficiency of resource utilization while minimizing the 652 possibility of congestion. 654 Increasingly, the Internet will have to function in the presence of 655 different classes of traffic with different service requirements. The 656 advent of Differentiated Services [RFC 2475] makes this requirement 657 particularly acute. Thus, packets may be grouped into behavior 658 aggregates such that each behavior aggregate may have a common set of 659 behavioral characteristics or a common set of delivery requirements. 660 In practice, the delivery requirements of a specific set of packets 661 may be specified explicitly or implicitly. Two of the most important 662 traffic delivery requirements are capacity constraints and QoS 663 constraints. 665 Capacity constraints can be expressed statistically as peak rates, 666 mean rates, burst sizes, or as some deterministic notion of effective 667 bandwidth. QoS requirements can be expressed in terms of (1) 668 integrity constraints such as packet loss and (2) in terms of 669 temporal constraints such as timing restrictions for the delivery of 670 each packet (delay) and timing restrictions for the delivery of 671 consecutive packets belonging to the same traffic stream (delay 672 variation). 674 2.3 Problem Context 676 Fundamental problems exist in association with the operation of a 677 network described by the simple model of the previous subsection. 678 This subsection reviews the problem context in relation to the 679 traffic engineering function. 681 The identification, abstraction, representation, and measurement of 682 network features relevant to traffic engineering is a significant 683 issue. 685 One particularly important class of problems concerns how to 686 explicitly formulate the problems that traffic engineering attempts 687 to solve, how to identify the requirements on the solution space, how 688 to specify the desirable features of good solutions, how to actually 689 solve the problems, and how to measure and characterize the 690 effectiveness of the solutions. 692 Another class of problems concerns how to measure and estimate 693 relevant network state parameters. Effective traffic engineering 694 relies on a good estimate of the offered traffic load as well as a 695 view of the underlying topology and associated resource constraints. 696 A network-wide view of the topology is also a must for offline 697 planning. 699 Still another class of problems concerns how to characterize the 700 state of the network and how to evaluate its performance under a 701 variety of scenarios. The performance evaluation problem is two-fold. 702 One aspect of this problem relates to the evaluation of the system 703 level performance of the network. The other aspect relates to the 704 evaluation of the resource level performance, which restricts 705 attention to the performance analysis of individual network 706 resources. In this memo, we refer to the system level characteristics 707 of the network as the "macro-states" and the resource level 708 characteristics as the "micro-states." The system level 709 characteristics are also known as the emergent properties of the 710 network as noted earlier. Correspondingly, we shall refer to the 711 traffic engineering schemes dealing with network performance 712 optimization at the systems level as "macro-TE" and the schemes that 713 optimize at the individual resource level as "micro-TE." Under 714 certain circumstances, the system level performance can be derived 715 from the resource level performance using appropriate rules of 716 composition, depending upon the particular performance measures of 717 interest. 719 Another fundamental class of problems concerns how to effectively 720 optimize network performance. Performance optimization may entail 721 translating solutions to specific traffic engineering problems into 722 network configurations. Optimization may also entail some degree of 723 resource management control, routing control, and/or capacity 724 augmentation. 726 As noted previously, congestion is an undesirable phenomena in 727 operational networks. Therefore, the next subsection addresses the 728 issue of congestion and its ramifications within the problem context 729 of Internet traffic engineering. 731 2.3.1 Congestion and its Ramifications 733 Congestion is one of the most significant problems in an operational 734 IP context. A network element is said to be congested if it 735 experiences sustained overload over an interval of time. Congestion 736 almost always results in degradation of service quality to end users. 737 Congestion control schemes can include demand side policies and 738 supply side policies. Demand side policies may restrict access to 739 congested resources and/or dynamically regulate the demand to 740 alleviate the overload situation. Supply side policies may expand or 741 augment network capacity to better accommodate offered traffic. 742 Supply side policies may also re-allocate network resources by 743 redistributing traffic over the infrastructure. Traffic 744 redistribution and resource re-allocation serve to increase the 745 'effective capacity' seen by the demand. 747 The emphasis of this memo is primarily on congestion management 748 schemes falling within the scope of the network, rather than on 749 congestion management systems dependent upon sensitivity and 750 adaptivity from end-systems. That is, the aspects that are considered 751 in this memo with respect to congestion management are those 752 solutions that can be provided by control entities operating on the 753 network and by the actions of network administrators and network 754 operations systems. 756 2.4 Solution Context 758 The solution context for Internet traffic engineering involves 759 analysis, evaluation of alternatives, and choice between alternative 760 courses of action. Generally the solution context is predicated on 761 making reasonable inferences about the current or future state of the 762 network, and subsequently making appropriate decisions that may 763 involve a preference between alternative sets of action. More 764 specifically, the solution context demands reasonable estimates of 765 traffic workload, characterization of network state, deriving 766 solutions to traffic engineering problems which may be implicitly or 767 explicitly formulated, and possibly instantiating a set of control 768 actions. Control actions may involve the manipulation of parameters 769 associated with routing, control over tactical capacity acquisition, 770 and control over the traffic management functions. 772 The following list of instruments may be applicable to the solution 773 context of Internet traffic engineering. 775 (1) A set of policies, objectives, and requirements (which may be 776 context dependent) for network performance evaluation and 777 performance optimization. 779 (2) A collection of online and possibly offline tools and mechanisms 780 for measurement, characterization, modeling, and control 781 of Internet traffic and control over the placement and allocation 782 of network resources, as well as control over the mapping or 783 distribution of traffic onto the infrastructure. 785 (3) A set of constraints on the operating environment, the network 786 protocols, and the traffic engineering system itself. 788 (4) A set of quantitative and qualitative techniques and 789 methodologies for abstracting, formulating, and 790 solving traffic engineering problems. 792 (5) A set of administrative control parameters which may be 793 manipulated through a Configuration Management (CM) system. 794 The CM system itself may include a configuration control 795 subsystem, a configuration repository, a configuration 796 accounting subsystem, and a configuration auditing subsystem. 798 (6) A set of guidelines for network performance evaluation, 799 performance optimization, and performance improvement. 801 Derivation of traffic characteristics through measurement and/or 802 estimation is very useful within the realm of the solution space for 803 traffic engineering. Traffic estimates can be derived from customer 804 subscription information, traffic projections, traffic models, and 805 from actual empirical measurements. The empirical measurements may be 806 performed at the traffic aggregate level or at the flow level in 807 order to derive traffic statistics at various levels of detail. 808 Measurements at the flow level or on small traffic aggregates may be 809 performed at edge nodes, where traffic enters and leaves the network. 810 Measurements at large traffic aggregate levels may be performed 811 within the core of the network where potentially numerous traffic 812 flows may be in transit concurrently. 814 To conduct performance studies and to support planning of existing 815 and future networks, a routing analysis may be performed to determine 816 the path(s) the routing protocols will choose for various traffic 817 demands, and to ascertain the utilization of network resources as 818 traffic is routed through the network. The routing analysis should 819 capture the selection of paths through the network, the assignment of 820 traffic across multiple feasible routes, and the multiplexing of IP 821 traffic over traffic trunks (if such constructs exists) and over the 822 underlying network infrastructure. A network topology model is a 823 necessity for routing analysis. A network topology model may be 824 extracted from network architecture documents, from network designs, 825 from information contained in router configuration files, from 826 routing databases, from routing tables, or from automated tools that 827 discover and depict network topology information. Topology 828 information may also be derived from servers that monitor network 829 state, and from servers that perform provisioning functions. 831 Routing in operational IP networks can be administratively controlled 832 at various levels of abstraction including the manipulation of BGP 833 attributes and manipulation of IGP metrics. For path oriented 834 technologies such as MPLS, routing can be further controlled by the 835 manipulation of relevant traffic engineering parameters, resource 836 parameters, and administrative policy constraints. Within the 837 context of MPLS, the path of an explicit label switched path (LSP) 838 can be computed and established in various ways including: (1) 839 manually, (2) automatically online using constraint-based routing 840 processes implemented on label switching routers, and (3) 841 automatically offline using constraint-based routing entities 842 implemented on external traffic engineering support systems. 844 2.4.1 Combating the Congestion Problem 846 Minimizing congestion is a significant aspect of Internet traffic 847 engineering. This subsection gives an overview of the general 848 approaches that have been used or proposed to combat congestion 849 problems. 851 Congestion management policies can be categorized based upon the 852 following criteria (see e.g., [YARE95] for a more detailed taxonomy 853 of congestion control schemes): (1) Response time scale which can be 854 characterized as long, medium, or short; (2) reactive versus 855 preventive which relates to congestion control and congestion 856 avoidance; and (3) supply side versus demand side congestion 857 management schemes. These aspects are discussed in the following 858 paragraphs. 860 (1) Congestion Management based on Response Time Scales 862 - Long (weeks to months): Capacity planning works over a relatively 863 long time scale to expand network capacity based on estimates or 864 forecasts of future traffic demand and traffic distribution. Since 865 router and link provisioning take time and are generally expensive, 866 these upgrades are typically carried out in the weeks-to-months or 867 even years time scale. 869 - Medium (minutes to days): Several control policies fall within the 870 medium time scale category. Examples include: (1) Adjusting IGP 871 and/or BGP parameters to route traffic away or towards certain 872 segments of the network; (2) Setting up and/or adjusting some 873 explicitly routed label switched paths (ER-LSPs) in MPLS networks to 874 route some traffic trunks away from possibly congested resources or 875 towards possibly more favorable routes; (3) re-configuring the 876 logical topology of the network to make it correlate more closely 877 with the spatial traffic distribution using for example some 878 underlying path-oriented technology such as MPLS LSPs, ATM PVCs, or 879 optical channel trails. Many of these adaptive medium time scale 880 response schemes rely on a measurement system that monitors changes 881 in traffic distribution, traffic shifts, and network resource 882 utilization and subsequently provides feedback to the online and/or 883 offline traffic engineering mechanisms and tools which employ this 884 feedback information to trigger certain control actions to occur 885 within the network. The traffic engineering mechanisms and tools can 886 be implemented in a distributed fashion or in a centralized fashion, 887 and may have a hierarchical structure or a flat structure. The 888 comparative merits of distributed and centralized control structures 889 for networks are well known. A centralized scheme may have global 890 visibility into the network state and may produce potentially more 891 optimal solutions. However, centralized schemes are prone to single 892 points of failure and may not scale as well as distributed schemes. 893 Moreover, the information utilized by a centralized scheme may be 894 stale and may not reflect the actual state of the network. It is not 895 an objective of this memo to make a recommendation between 896 distributed and centralized schemes. This is a choice that network 897 administrators must make based on their specific needs. 899 - Short (picoseconds to minutes): This category includes packet level 900 processing functions and events on the order of several round trip 901 times. It includes router mechanisms such as passive and active 902 buffer management. These mechanisms are used to control congestion 903 and/or signal congestion to end systems so that they can adaptively 904 regulate the rate at which traffic is injected into the network. One 905 of the most popular active queue management schemes, especially for 906 TCP traffic, is Random Early Detection (RED) [FLJA93], which supports 907 congestion avoidance by controlling the average queue size. During 908 congestion (but before the queue is filled), the RED scheme chooses 909 arriving packets to "mark" according to a probabilistic algorithm 910 which takes into account the average queue size. For a router that 911 does not utilize explicit congestion notification (ECN) see e.g., 912 [FLOY94], the marked packets can simply be dropped to signal the 913 inception of congestion to end systems. On the other hand, if the 914 router supports ECN, then it can set the ECN field in the packet 915 header. Several variations of RED have been proposed to support 916 different drop precedence levels in multi-class environments [RFC- 917 2597], e.g., RED with In and Out (RIO) and Weighted RED. There is 918 general consensus that RED provides congestion avoidance performance 919 which is not worse than traditional Tail-Drop (TD) queue management 920 (drop arriving packets only when the queue is full). Importantly, 921 however, RED reduces the possibility of global synchronization and 922 improves fairness among different TCP sessions. However, RED by 923 itself can not prevent congestion and unfairness caused by sources 924 unresponsive to RED, e.g., UDP traffic and some misbehaved greedy 925 connections. Other schemes have been proposed to improve the 926 performance and fairness in the presence of unresponsive traffic. 927 Some of these schemes were proposed as theoretical frameworks and are 928 typically not available in existing commercial products. Two such 929 schemes are Longest Queue Drop (LQD) and Dynamic Soft Partitioning 930 with Random Drop (RND) [SLDC98]. 932 (2) Congestion Management: Reactive versus Preventive Schemes 934 - Reactive: reactive (recovery) congestion management policies react 935 to existing congestion problems to improve it. All the policies 936 described in the long and medium time scales above can be categorized 937 as being reactive especially if the policies are based on monitoring 938 and identifying existing congestion problems, and on the initiation 939 of relevant actions to ease a situation. 941 - Preventive: preventive (predictive/avoidance) policies take 942 proactive action to prevent congestion based on estimates and 943 predictions of future potential congestion problems. Some of the 944 policies described in the long and medium time scales fall into this 945 category. They do not necessarily respond immediately to existing 946 congestion problems. Instead forecasts of traffic demand and workload 947 distribution are considered and action may be taken to prevent 948 potential congestion problems in the future. The schemes described in 949 the short time scale (e.g., RED and its variations, ECN, LQD, and 950 RND) are also used for congestion avoidance since dropping or marking 951 packets before queues actually overflow would trigger corresponding 952 TCP sources to slow down. 954 (3) Congestion Management: Supply Side versus Demand Side Schemes 956 - Supply side: supply side congestion management policies increase 957 the effective capacity available to traffic in order to control or 958 obviate congestion. This can be accomplished by augmenting capacity. 959 Another way to accomplish this is to minimize congestion by having a 960 relatively balanced distribution of traffic over the network. For 961 example, capacity planning should aim to provide a physical topology 962 and associated link bandwidths that match estimated traffic workload 963 and traffic distribution based on forecasting (subject to budgetary 964 and other constraints). However, if actual traffic distribution does 965 not match the topology derived from capacity panning (due to 966 forecasting errors or facility constraints for example), then the 967 traffic can be mapped onto the existing topology using routing 968 control mechanisms, using path oriented technologies (e.g., MPLS LSPs 969 and optical channel trails) to modify the logical topology, or by 970 using some other load redistribution mechanisms. 972 - Demand side: demand side congestion management policies control or 973 regulate the offered traffic to alleviate congestion problems. For 974 example, some of the short time scale mechanisms described earlier 975 (such as RED and its variations, ECN, LQD, and RND) as well as 976 policing and rate shaping mechanisms attempt to regulate the offered 977 load in various ways. Tariffs may also be applied as a demand side 978 instrument. To date, however, tariffs have not been used as a means 979 of demand side congestion management within the Internet. 981 In summary, a variety of mechanisms can be used to address congestion 982 problems in IP networks. These mechanisms may operate at multiple 983 time-scales. 985 2.5 Implementation and Operational Context 987 The operational context of Internet traffic engineering is 988 characterized by constant change which occur at multiple levels of 989 abstraction. The implementation context demands effective planning, 990 organization, and execution. The planning aspects may involve 991 determining prior sets of actions to achieve desired objectives. 992 Organizing involves arranging and assigning responsibility to the 993 various components of the traffic engineering system and coordinating 994 the activities to accomplish the desired TE objectives. Execution 995 involves measuring and applying corrective or perfective actions to 996 attain and maintain desired TE goals. 998 3.0 Traffic Engineering Process Model(s) 1000 This section describes a generic process model that captures the high 1001 level practical aspects of Internet traffic engineering in an 1002 operational context. The process model is described as a sequence of 1003 actions that a traffic engineer, or more generally a traffic 1004 engineering system, must perform to optimize the performance of an 1005 operational network (see also [RFC-2702, AWD2]). The process model 1006 described here represents the broad activities common to most traffic 1007 engineering methodologies although the details regarding how traffic 1008 engineering is executed may differ from network to network. This 1009 process model may be enacted explicitly or implicitly, by an 1010 automaton and/or by a human. 1012 The traffic engineering process model is iterative [AWD2]. The four 1013 phases of the process model described below are repeated continually. 1015 The first phase of the TE process model is to define the relevant 1016 control policies that govern the operation of the network. These 1017 policies may depend upon many factors including the prevailing 1018 business model, the network cost structure, the operating 1019 constraints, the utility model, and optimization criteria. 1021 The second phase of the process model is a feedback mechanism 1022 involving the acquisition of measurement data from the operational 1023 network. If empirical data is not readily available from the network, 1024 then synthetic workloads may be used instead which reflect either the 1025 prevailing or the expected workload of the network. Synthetic 1026 workloads may be derived by estimation or extrapolation using prior 1027 empirical data. Their derivation may also be obtained using 1028 mathematical models of traffic characteristics or other means. 1030 The third phase of the process model is to analyze the network state 1031 and to characterize traffic workload. Performance analysis may be 1032 proactive and/or reactive. Proactive performance analysis identifies 1033 potential problems that do not exist, but could manifest in the 1034 future. Reactive performance analysis identifies existing problems, 1035 determines their cause through diagnosis, and evaluates alternative 1036 approaches to remedy the problem, if necessary. A number of 1037 quantitative and qualitative techniques may be used in the analysis 1038 process, including modeling based analysis and simulation. The 1039 analysis phase of the process model may involve investigating the 1040 concentration and distribution of traffic across the network or 1041 relevant subsets of the network, identifying the characteristics of 1042 the offered traffic workload, identifying existing or potential 1043 bottlenecks, and identifying network pathologies such as ineffective 1044 link placement, single points of failures, etc. Network pathologies 1045 may result from many factors including inferior network architecture, 1046 inferior network design, and configuration problems. A traffic 1047 matrix may be constructed as part of the analysis process. Network 1048 analysis may also be descriptive or prescriptive. 1050 The fourth phase of the TE process model is the performance 1051 optimization of the network. The performance optimization phase 1052 involves a decision process which selects and implements a set of 1053 actions from a set of alternatives. Optimization actions may include 1054 the use of appropriate techniques to either control the offered 1055 traffic or to control the distribution of traffic across the network. 1056 Optimization actions may also involve adding additional links or 1057 increasing link capacity, deploying additional hardware such as 1058 routers and switches, systematically adjusting parameters associated 1059 with routing such as IGP metrics and BGP attributes, and adjusting 1060 traffic management parameters. Network performance optimization may 1061 also involve starting a network planning process to improve the 1062 network architecture, network design, network capacity, network 1063 technology, and the configuration of network elements to accommodate 1064 current and future growth. 1066 3.1 Components of the Traffic Engineering Process Model 1068 The key components of the traffic engineering process model include a 1069 measurement subsystem, a modeling and analysis subsystem, and an 1070 optimization subsystem. The following subsections examine these 1071 components as they apply to the traffic engineering process model. 1073 3.2 Measurement 1075 Measurement is crucial to the traffic engineering function. The 1076 operational state of a network can be conclusively determined only 1077 through measurement. Measurement is also critical to the optimization 1078 function because it provides feedback data which is used by traffic 1079 engineering control subsystems. This data is used to adaptively 1080 optimize network performance in response to events and stimuli 1081 originating within and outside the network. Measurement is also 1082 needed to determine the quality of network services and to evaluate 1083 the effectiveness of traffic engineering policies. Experience 1084 suggests that measurement is most effective when acquired and applied 1085 systematically. 1087 When developing a measurement system to support the traffic 1088 engineering function in IP networks, the following questions should 1089 be carefully considered: Why is measurement needed in this particular 1090 context? What parameters are to be measured? How should the 1091 measurement be accomplished? Where should the measurement be 1092 performed? When should the measurement be performed? How frequently 1093 should the monitored variables be measured? What level of 1094 measurement accuracy and reliability is desirable? What level of 1095 measurement accuracy and reliability is realistically attainable? To 1096 what extent can the measurement system permissibly interfere with the 1097 monitored network components and variables? What is the acceptable 1098 cost of measurement? The answers to these questions will determine 1099 the measurement tools and methodologies appropriate in any given 1100 traffic engineering context. 1102 It should also be noted that there is a distinction between 1103 measurement and evaluation. Measurement provides raw data concerning 1104 state parameters and variables of monitored network elements. 1105 Evaluation utilizes the raw data to make inferences regarding the 1106 monitored system. 1108 Measurement in support of the TE function can occur at different 1109 levels of abstraction. For example, measurement can be used to derive 1110 packet level characteristics, flow level characteristics, user or 1111 customer level characteristics, traffic aggregate characteristics, 1112 component level characteristics, and network wide characteristics. 1114 3.3 Modeling, Analysis, and Simulation 1116 Modeling and analysis are important aspects of Internet traffic 1117 engineering. Modeling involves constructing an abstract or physical 1118 representation which depicts relevant traffic characteristics and 1119 network attributes. 1121 A network model is an abstract representation of the network which 1122 captures relevant network features, attributes, and characteristics, 1123 such as link and nodal attributes and constraints. A network model 1124 may facilitate analysis and/or simulation which can be used to 1125 predict network performance under various conditions as well as to 1126 guide network expansion plans. 1128 In general, Internet traffic engineering models can be classified as 1129 either structural or behavioral. Structural models focus on the 1130 organization of the network and its components. Behavioral models 1131 focus on the dynamics of the network and the traffic workload. 1132 Modeling for Internet traffic engineering may also be formal or 1133 informal. 1135 Accurate behavioral models for traffic sources are particularly 1136 useful for analysis. Development of behavioral traffic source models 1137 that are consistent with empirical data obtained from operational 1138 networks is a major research topic in Internet traffic engineering. 1139 These source models should also be tractable and amenable to 1140 analysis. The topic of source models for IP traffic is a research 1141 topic and is therefore outside the scope of this document. Its 1142 importance, however, must be emphasized. 1144 Network simulation tools are extremely useful for traffic 1145 engineering. Because of the complexity of realistic quantitative 1146 analysis of network behavior, certain aspects of network performance 1147 studies can only be conducted effectively using simulation. A good 1148 network simulator can be used to mimic and visualize network 1149 characteristics under various conditions in a safe and non-disruptive 1150 manner. For example, a network simulator may be used to depict 1151 congested resources and hot spots, and to provide hints regarding 1152 possible solutions to network performance problems. A good simulator 1153 may also be used to validate the effectiveness of planned solutions 1154 to network issues without the need to tamper with the operational 1155 network, or to commence an expensive network upgrade which may not 1156 achieve the desired objectives. Furthermore, during the process of 1157 network planning, a network simulator may reveal pathologies such as 1158 single points of failure which may require additional redundancy, and 1159 potential bottlenecks and hot spots which may require additional 1160 capacity. 1162 Routing simulators are especially useful in large networks. A routing 1163 simulator may identify planned links which may not actually be used 1164 to route traffic by the existing routing protocols. Simulators can 1165 also be used to conduct scenario based and perturbation based 1166 analysis, as well as sensitivity studies. Simulation results can be 1167 used to initiate appropriate actions in various ways. For example, an 1168 important application of network simulation tools is to investigate 1169 and identify how best to evolve and grow the network in order to 1170 accommodate projected future demands. 1172 3.4 Optimization 1174 Network performance optimization involves resolving network issues by 1175 transforming such issues into concepts that enable a solution, 1176 identification of a solution, and implementation of the solution. 1177 Network performance optimization can be corrective or perfective. In 1178 corrective optimization, the goal is to remedy a problem that has 1179 occurred or that is incipient. In perfective optimization, the goal 1180 is to improve network performance even when explicit problems do not 1181 exist and are not anticipated. 1183 Network performance optimization is a continual process, as noted 1184 previously. Performance optimization iterations may consist of 1185 real-time optimization sub-processes and non-real-time network 1186 planning sub-processes. The difference between real-time 1187 optimization and network planning is primarily in the relative time- 1188 scale in they operate and in the granularity of actions. One of the 1189 objectives of a real-time optimization sub-process is to control the 1190 mapping and distribution of traffic over the existing network 1191 infrastructure to avoid and/or relieve congestion, to assure 1192 satisfactory service delivery, and to optimize resource utilization. 1193 Real-time optimization is needed because random incidents such as 1194 fiber cuts or shifts in traffic demand will occur irrespective of how 1195 well a network is designed. These incidents can cause congestion and 1196 other problems to manifest in an operational network. Real-time 1197 optimization must solve such problems in small to medium time-scales 1198 ranging from micro-seconds to minutes or hours. Examples of real-time 1199 optimization include queue management, IGP/BGP metric tuning, and 1200 using technologies such as MPLS explicit LSPs to change the paths of 1201 some traffic trunks [XIAO]. 1203 One of the functions of the network planning sub-process is to 1204 initiate actions to systematically evolve the architecture, 1205 technology, topology, and capacity of a network. When a problem 1206 exists in the network, real-time optimization should provide an 1207 immediate remedy. Because a prompt response is necessary, the real- 1208 time solution may not be the best possible solution. Network 1209 planning may subsequently be needed to refine the solution and 1210 improve the situation. Network planning is also required to expand 1211 the network to support traffic growth and changes in traffic 1212 distribution over time. As previously noted, a change in the topology 1213 and/or capacity of the network may be the outcome of network 1214 planning. 1216 Clearly, network planning and real-time performance optimization are 1217 mutually complementary activities. A well-planned and designed 1218 network makes real-time optimization easier, while a systematic 1219 approach to real-time network performance optimization allows network 1220 planning to focus on long term issues rather than tactical 1221 considerations. Systematic real-time network performance 1222 optimization also provides valuable inputs and insights toward 1223 network planning. 1225 Stability is an important consideration in real-time network 1226 performance optimization. This aspect will be repeatedly addressed 1227 throughout this memo. 1229 4.0 Historical Review and Recent Developments 1231 This section briefly reviews different traffic engineering approaches 1232 proposed and implemented in telecommunications and computer networks. 1233 The discussion is not intended to be comprehensive. It is primarily 1234 intended to illuminate pre-existing perspectives and prior art 1235 concerning traffic engineering in the Internet and in legacy 1236 telecommunications networks. 1238 4.1 Traffic Engineering in Classical Telephone Networks 1240 This subsection presents a brief overview of traffic engineering in 1241 telephone networks which often relates to the way user traffic is 1242 steered from an originating node to the terminating node. This 1243 subsection presents a brief overview of this topic. A detailed 1244 description of the various routing strategies applied in telephone 1245 networks is included in the book by G. Ash [ASH2]. 1247 The early telephone network relied on static hierarchical routing, 1248 whereby routing patterns remained fixed independent of the state of 1249 the network or time of day. The hierarchy was intended to accommodate 1250 overflow traffic, improve network reliability via alternate routes, 1251 and prevent call looping by employing strict hierarchical rules. The 1252 network was typically over-provisioned since a given fixed route had 1253 to be dimensioned so that it could carry user traffic during a busy 1254 hour of any busy day. Hierarchical routing in the telephony network 1255 was found to be too rigid upon the advent of digital switches and 1256 stored program control which were able to manage more complicated 1257 traffic engineering rules. 1259 Dynamic routing was introduced to alleviate the routing inflexibility 1260 in the static hierarchical routing so that the network would operate 1261 more efficiently. This resulted in significant economic gains 1262 [HUSS87]. Dynamic routing typically reduces the overall loss 1263 probability by 10 to 20 percent (compared to static hierarchical 1264 routing). Dynamic routing can also improve network resilience by 1265 recalculating routes on a per-call basis and periodically updating 1266 routes. 1268 There are three main types of dynamic routing in the telephone 1269 network. They are time-dependent routing, state-dependent routing 1270 (SDR), and event dependent routing (EDR). 1272 In time-dependent routing, regular variations in traffic loads (such 1273 as time of day or day of week) are exploited in pre-planned routing 1274 tables. In state-dependent routing, routing tables are updated 1275 online according to the current state of the network (e.g, traffic 1276 demand, utilization, etc.). In event dependent routing, routing 1277 changes are incepted by events (such as call setups encountering 1278 congested or blocked links) whereupon new paths are searched out 1279 using learning models. EDR methods are real-time adaptive, but they 1280 do not require global state information as does SDR. Examples of EDR 1281 schemes include the dynamic alternate routing (DAR) from BT, the 1282 state-and-time dependent routing (STR) from NTT, and the success-to- 1283 the-top (STT) routing from AT&T. 1285 Dynamic non-hierarchical routing (DNHR) is an example of dynamic 1286 routing that was introduced in the AT&T toll network in the 1980's to 1287 respond to time-dependent information such as regular load variations 1288 as a function of time. Time-dependent information in terms of load 1289 may be divided into three time scales: hourly, weekly, and yearly. 1290 Correspondingly, three algorithms are defined to pre-plan the routing 1291 tables. The network design algorithm operates over a year-long 1292 interval while the demand servicing algorithm operates on a weekly 1293 basis to fine tune link sizes and routing tables to correct forecast 1294 errors on the yearly basis. At the smallest time scale, the routing 1295 algorithm is used to make limited adjustments based on daily traffic 1296 variations. Network design and demand servicing are computed using 1297 offline calculations. Typically, the calculations require extensive 1298 search on possible routes. On the other hand, routing may need 1299 online calculations to handle crankback. DNHR adopts a "two-link" 1300 approach whereby a path can consist of two links at most. The 1301 routing algorithm presents an ordered list of route choices between 1302 an originating switch and a terminating switch. If a call overflows, 1303 a via switch (a tandem exchange between the originating switch and 1304 the terminating switch) would send a crankback signal to the 1305 originating switch. This switch would then select the next route, 1306 and so on, until there are no alternative routes available in which 1307 the call is blocked. 1309 4.2 Evolution of Traffic Engineering in Packet Networks 1311 This subsection reviews related prior work that was intended to 1312 improve the performance of data networks. Indeed, optimization of 1313 the performance of data networks started in the early days of the 1314 ARPANET. Other early commercial networks such as SNA also recognized 1315 the importance of performance optimization and service 1316 differentiation. 1318 In terms of traffic management, the Internet has been a best effort 1319 service environment until recently. In particular, very limited 1320 traffic management capabilities existed in IP networks to provide 1321 differentiated queue management and scheduling services to packets 1322 belonging to different classes. 1324 In terms of routing control, the Internet has employed distributed 1325 protocols for intra-domain routing. These protocols are highly 1326 scalable and resilient. However, they are based on simple algorithms 1327 for path selection which have very limited functionality to allow 1328 flexible control of the path selection process. 1330 In the following subsections, the evolution of practical traffic 1331 engineering mechanisms in IP networks and its predecessors is 1332 reviewed. 1334 4.2.1 Adaptive Routing in the ARPANET 1336 The early ARPANET recognized the importance of adaptive routing where 1337 routing decisions were based on the current state of the network 1338 [MCQ80]. Early minimum delay routing approaches forwarded each 1339 packet to its destination along a path for which the total estimated 1340 transit time was the smallest. Each node maintained a table of 1341 network delays, representing the estimated delay that a packet would 1342 experience along a given path toward its destination. The minimum 1343 delay table was periodically transmitted by a node to its neighbors. 1344 The shortest path, in terms of hop count, was also propagated to give 1345 the connectivity information. 1347 One drawback to this approach is that dynamic link metrics tend to 1348 create "traffic magnets" causing congestion to be shifted from one 1349 location of a network to another location, resulting in oscillation 1350 and network instability. 1352 4.2.2 Dynamic Routing in the Internet 1354 The Internet evolved from the APARNET and adopted dynamic routing 1355 algorithms with distributed control to determine the paths that 1356 packets should take en-route to their destinations. The routing 1357 algorithms are adaptations of shortest path algorithms where costs 1358 are based on link metrics. The link metric can be based on static or 1359 dynamic quantities. The link metric based on static quantities may be 1360 assigned administratively according to local criteria. The link 1361 metric based on dynamic quantities may be a function of a network 1362 congestion measure such as delay or packet loss. 1364 It was apparent early that static link metric assignment was 1365 inadequate because it can easily lead to unfavorable scenarios in 1366 which some links become congested while others remain lightly loaded. 1367 One of the many reasons for the inadequacy of static link metrics is 1368 that link metric assignment was often done without considering the 1369 traffic matrix in the network. Also, the routing protocols did not 1370 take traffic attributes and capacity constraints into account when 1371 making routing decisions. This results in traffic concentration being 1372 localized in subsets of the network infrastructure and potentially 1373 causing congestion. Even if link metrics are assigned in accordance 1374 with the traffic matrix, unbalanced loads in the network can still 1375 occur due to a number factors including: 1377 - Resources may not be deployed in the most optimal locations 1378 from a routing perspective. 1380 - Forecasting errors in traffic volume and/or traffic distribution. 1382 - Dynamics in traffic matrix due to the temporal nature of traffic 1383 patterns, BGP policy change from peers, etc. 1385 The inadequacy of the legacy Internet interior gateway routing system 1386 is one of the factors motivating the interest in path oriented 1387 technology with explicit routing and constraint-based routing 1388 capability such as MPLS. 1390 4.2.3 ToS Routing 1392 Type-of-Service (ToS) routing involves different routes going to the 1393 same destination being selected depending upon the ToS field of an IP 1394 packet [RFC-1349]. The ToS classes may be classified as low delay 1395 and high throughput. Each link is associated with multiple link 1396 costs and each link cost is used to compute routes for a particular 1397 ToS. A separate shortest path tree is computed for each ToS. The 1398 shortest path algorithm must be run for each ToS resulting in very 1399 expensive computation. Classical ToS-based routing is now outdated 1400 as the IP header field has been replaced by a Diffserv field. 1401 Effective traffic engineering is difficult to perform in classical 1402 ToS-based routing because each class still relies exclusively on 1403 shortest path routing which results in localization of traffic 1404 concentration within the network. 1406 4.2.4 Equal Cost Multi-Path 1408 Equal Cost Multi-Path (ECMP) is another technique that attempts to 1409 address the deficiency in Shortest Path First (SPF) interior gateway 1410 routing systems [RFC-2178]. In the classical SPF algorithm, if two or 1411 more shortest paths exist to a given destination, the algorithm will 1412 choose one of them. The algorithm is modified slightly in ECMP so 1413 that if two or more equal cost shortest paths exist between two 1414 nodes, the traffic between the nodes is distributed among the 1415 multiple equal-cost paths. Traffic distribution across the equal- 1416 cost paths is usually performed in one of two ways: (1) packet-based 1417 in a round-robin fashion, or (2) flow-based using hashing on source 1418 and destination IP addresses and possibly other fields of the IP 1419 header. The first approach can easily cause out-of-order packets 1420 while the second approach is dependent upon the number and 1421 distribution of flows. Flow-based load sharing may be unpredictable 1422 in an enterprise network where the number of flows is relatively 1423 small and less heterogeneous (for example, hashing may not be 1424 uniform), but it is generally effective in core public networks where 1425 the number of flows is large and heterogeneous. 1427 In ECMP, link costs are static and bandwidth constraints are not 1428 considered, so ECMP attempts to distribute the traffic as equally as 1429 possible among the equal-cost paths independent of the congestion 1430 status of each path. As a result, given two equal-cost paths, it is 1431 possible that one of the paths will be more congested than the other. 1432 Another drawback of ECMP is that load sharing cannot be achieved on 1433 multiple paths which have non-identical costs. 1435 4.2.5 Nimrod 1437 Nimrod is a routing system developed to provide heterogeneous service 1438 specific routing in the Internet, while taking multiple constraints 1439 into account [RFC-1992]. Essentially, Nimrod is a link state routing 1440 protocol which supports path oriented packet forwarding. It uses the 1441 concept of maps to represent network connectivity and services at 1442 multiple levels of abstraction. Mechanisms are provided to allow 1443 restriction of the distribution of routing information. 1445 Even though Nimrod did not enjoy deployment in the public Internet, a 1446 number of key concepts incorporated into the Nimrod architecture, 1447 such as explicit routing which allows selection of paths at 1448 originating nodes, are beginning to find applications in some recent 1449 constraint-based routing initiatives. 1451 4.3 Overlay Model 1453 In the overlay model, a virtual-circuit network, such as ATM, frame 1454 relay, or WDM provides virtual-circuit connectivity between routers 1455 that are located at the edges of a virtual-circuit cloud. In this 1456 mode, two routers that are connected through a virtual circuit see a 1457 direct adjacency between themselves independent of the physical route 1458 taken by the virtual circuit through the ATM, frame relay, or WDM 1459 network. Thus, the overlay model essentially decouples the logical 1460 topology that routers see from the physical topology that the ATM, 1461 frame relay, or WDM network manages. The overlay model based on ATM 1462 or frame relay enables a network administrator or an automaton to 1463 employ traffic engineering concepts to perform path optimization by 1464 re-configuring or rearranging the virtual circuits so that a virtual 1465 circuit on a congested or sub-optimal physical link can be re-routed 1466 to a less congested or more optimal one. In the overlay model, 1467 traffic engineering is also employed to establish relationships 1468 between the traffic management parameters (e.g. PCR, SCR, and MBS for 1469 ATM) of the virtual-circuit technology and the actual traffic that 1470 traverses each circuit. These relationships can be established based 1471 upon known or projected traffic profiles, and some other factors. 1473 The overlay model using IP over ATM requires the management of two 1474 separate networks with different technologies (IP and ATM) resulting 1475 in increased operational complexity and cost. In the fully-meshed 1476 overlay model, each router would peer to every other router in the 1477 network, so that the total number of adjacencies is a quadratic 1478 function of the number of routers. Some of the issues with the 1479 overlay model are discussed in [AWD2]. 1481 4.4 Constrained-Based Routing 1483 Constraint-based routing refers to a class of routing systems that 1484 compute routes through a network subject to satisfaction of a set of 1485 constraints and requirements. In the most general setting, 1486 constraint-based routing may also seek to optimize overall network 1487 performance while minimizing costs. 1489 The constraints and requirements may be imposed by the network itself 1490 or by administrative policies. Constraints may include bandwidth, hop 1491 count, delay, and policy instruments such as resource class 1492 attributes. Constraints may also include domain specific attributes 1493 of certain network technologies and contexts which impose 1494 restrictions on the solution space of the routing function. Path 1495 oriented technologies such as MPLS have made constraint-based routing 1496 feasible and attractive in public IP networks. 1498 The concept of constraint-based routing within the context of MPLS 1499 traffic engineering requirements in IP networks was first defined in 1500 [RFC-2702]. 1502 Unlike QoS routing (for example, see [RFC-2386] and [MA]) which 1503 generally addresses the issue of routing individual traffic flows to 1504 satisfy prescribed flow based QoS requirements subject to network 1505 resource availability, constraint-based routing is applicable to 1506 traffic aggregates as well as flows and may be subject to a wide 1507 variety of constraints which may include policy restrictions. 1509 4.5 Overview of Other IETF Projects Related to Traffic Engineering 1511 This subsection reviews a number of IETF activities pertinent to 1512 Internet traffic engineering. These activities are primarily intended 1513 to evolve the IP architecture to support new service definitions 1514 which allow preferential or differentiated treatment to be accorded 1515 to certain types of traffic. 1517 4.5.1 Integrated Services 1519 The IETF Integrated Services working group developed the integrated 1520 services (Intserv) model. This model requires resources, such as 1521 bandwidth and buffers, to be reserved a priori for a given traffic 1522 flow to ensure that the quality of service requested by the traffic 1523 flow is satisfied. The integrated services model includes additional 1524 components beyond those used in the best-effort model such as packet 1525 classifiers, packet schedulers, and admission control. A packet 1526 classifier is used to identify flows that are to receive a certain 1527 level of service. A packet scheduler handles the scheduling of 1528 service to different packet flows to ensure that QoS commitments are 1529 met. Admission control is used to determine whether a router has the 1530 necessary resources to accept a new flow. 1532 Two services have been defined under the Integrated Services model: 1533 guaranteed service [RFC-2212] and controlled-load service [RFC-2211]. 1535 The guaranteed service can be used for applications requiring bounded 1536 packet delivery time. For this type of application, data that is 1537 delivered to the application after a pre-defined amount of time has 1538 elapsed is usually considered worthless. Therefore, guaranteed 1539 service was intended to provide a firm quantitative bound on the 1540 end-to-end packet delay for a flow. This is accomplished by 1541 controlling the queuing delay on network elements along the data flow 1542 path. The guaranteed service model does not, however, provide bounds 1543 on jitter (inter-arrival times between consecutive packets). 1545 The controlled-load service can be used for adaptive applications 1546 that can tolerate some delay but are sensitive to traffic overload 1547 conditions. This type of application typically functions 1548 satisfactorily when the network is lightly loaded but its performance 1549 degrades significantly when the network is heavily loaded. 1550 Controlled-load service therefore has been designed to provide 1551 approximately the same service as best-effort service in a lightly 1552 loaded network regardless of actual network conditions. Controlled- 1553 load service is described qualitatively in that no target values of 1554 delay or loss are specified. 1556 The main issue with the Integrated Services model has been 1557 scalability [RFC-2998], especially in large public IP networks which 1558 may potentially have millions of active micro-flows in transit 1559 concurrently. 1561 A notable feature of the Integrated Services model is that it 1562 requires explicit signaling of QoS requirements from end systems to 1563 routers [RFC-2753]. The Resource Reservation Protocol (RSVP) performs 1564 this signaling function and is a critical component of the Integrated 1565 Services model. The RSVP protocol is described next. 1567 4.5.2 RSVP 1569 RSVP is a soft state signaling protocol [RFC-2205]. It supports 1570 receiver initiated establishment of resource reservations for both 1571 multicast and unicast flows. RSVP was originally developed as a 1572 signaling protocol within the integrated services framework for 1573 applications to communicate QoS requirements to the network and for 1574 the network to reserve relevant resources to satisfy the QoS 1575 requirements [RFC-2205]. 1577 Under RSVP, the sender or source node sends a PATH message to the 1578 receiver with the same source and destination addresses as the 1579 traffic which the sender will generate. The PATH message contains: 1580 (1) a sender Tspec specifying the characteristics of the traffic, (2) 1581 a sender Template specifying the format of the traffic, and (3) an 1582 optional Adspec which is used to support the concept of one pass with 1583 advertising" (OPWA) [RFC-2205]. Every intermediate router along the 1584 path forwards the PATH Message to the next hop determined by the 1585 routing protocol. Upon receiving a PATH Message, the receiver 1586 responds with a RESV message which includes a flow descriptor used to 1587 request resource reservations. The RESV message travels to the sender 1588 or source node in the opposite direction along the path that the PATH 1589 message traversed. Every intermediate router along the path can 1590 reject or accept the reservation request of the RESV message. If the 1591 request is rejected, the rejecting router will send an error message 1592 to the receiver and the signaling process will terminate. If the 1593 request is accepted, link bandwidth and buffer space are allocated 1594 for the flow and the related flow state information is installed in 1595 the router. 1597 One of the issues with the original RSVP specification was 1598 Scalability. This is because reservations were required for micro- 1599 flows, so that the amount of state maintained by network elements 1600 tends to increase linearly with the number of micro-flows. These 1601 issues are described in [RFC-2961]. 1603 Recently, RSVP has been modified and extended in several ways to 1604 mitigate the scaling problems. As a result, it is becoming a 1605 versatile signaling protocol for the Internet. For example, RSVP has 1606 been extended to reserve resources for aggregation of flows, to set 1607 up MPLS explicit label switched paths, and to perform other signaling 1608 functions within the Internet. There are also a number of proposals 1609 to reduce the amount of refresh messages required to maintain 1610 established RSVP sessions [RFC-2961]. 1612 A number of IETF working groups have been engaged in activities 1613 related to the RSVP protocol. These include the original RSVP working 1614 group, the MPLS working group, the Resource Allocation Protocol 1615 working group, and the Policy Framework working group. 1617 4.5.3 Differentiated Services 1619 The goal of the Differentiated Services (Diffserv) effort within the 1620 IETF is to devise scalable mechanisms for categorization of traffic 1621 into behavior aggregates, which ultimately allows each behavior 1622 aggregate to be treated differently, especially when there is a 1623 shortage of resources such as link bandwidth and buffer space [RFC- 1624 2475]. One of the primary motivations for the Diffserv effort was to 1625 devise alternative mechanisms for service differentiation in the 1626 Internet that mitigate the scalability issues encountered with the 1627 Intserv model. 1629 The IETF Diffserv working group has defined a Differentiated Services 1630 field in the IP header (DS field). The DS field consists of six bits 1631 of the part of the IP header formerly known as TOS octet. The DS 1632 field is used to indicate the forwarding treatment that a packet 1633 should receive at a node [RFC-2474]. The Diffserv working group has 1634 also standardized a number of Per-Hop Behavior (PHB) groups. Using 1635 the PHBs, several classes of services can be defined using different 1636 classification, policing, shaping and scheduling rules. 1638 For an end-user of network services to receive Differentiated 1639 Services from its Internet Service Provider (ISP), it may be 1640 necessary for the user to have a Service Level Agreement (SLA) with 1641 the ISP. An SLA may explicitly or implicitly specify a Traffic 1642 Conditioning Agreement (TCA) which defines classifier rules as well 1643 as metering, marking, discarding, and shaping rules. 1645 Packets are classified, and possibly policed and shaped at the 1646 ingress to a Diffserv network. When a packet traverses the boundary 1647 between different Diffserv domains, the DS field of the packet may be 1648 re-marked according to existing agreements between the domains. 1650 Differentiated Services allows only a finite number of service 1651 classes to be indicated by the DS field. The main advantage of the 1652 Diffserv approach relative to the Intserv model is scalability. 1653 Resources are allocated on a per-class basis and the amount of state 1654 information is proportional to the number of classes rather than to 1655 the number of application flows. 1657 It should be obvious from the previous discussion that the Diffserv 1658 model essentially deals with traffic management issues on a per hop 1659 basis. The Diffserv control model consists of a collection of micro- 1660 TE control mechanisms. Other traffic engineering capabilities, such 1661 as capacity management (including routing control), are also required 1662 in order to deliver acceptable service quality in Diffserv networks. 1663 The concept of Per Domain Behaviors has been introduced to better 1664 capture the notion of differentiated services across a complete 1665 domain [RFC-3086]. 1667 4.5.4 MPLS 1669 MPLS is an advanced forwarding scheme which also includes extensions 1670 to conventional IP control plane protocols. MPLS extends the Internet 1671 routing model and enhances packet forwarding and path control [RFC- 1672 3031]. 1674 At the ingress to an MPLS domain, label switching routers (LSRs) 1675 classify IP packets into forwarding equivalence classes (FECs) based 1676 on a variety of factors, including e.g. a combination of the 1677 information carried in the IP header of the packets and the local 1678 routing information maintained by the LSRs. An MPLS label is then 1679 prepended to each packet according to their forwarding equivalence 1680 classes. In a non-ATM/FR environment, the label is 32 bits long and 1681 contains a 20-bit label field, a 3-bit experimental field (formerly 1682 known as Class-of-Service or CoS field), a 1-bit label stack 1683 indicator and an 8-bit TTL field. In an ATM (FR) environment, the 1684 label consists information encoded in the VCI/VPI (DLCI) field. An 1685 MPLS capable router (an LSR) examines the label and possibly the 1686 experimental field and uses this information to make packet 1687 forwarding decisions. 1689 An LSR makes forwarding decisions by using the label prepended to 1690 packets as the index into a local next hop label forwarding entry 1691 (NHLFE). The packet is then processed as specified in the NHLFE. The 1692 incoming label may be replaced by an outgoing label, and the packet 1693 may be switched to the next LSR. This label-switching process is very 1694 similar to the label (VCI/VPI) swapping process in ATM networks. 1695 Before a packet leaves an MPLS domain, its MPLS label may be removed. 1696 A Label Switched Path (LSP) is the path between an ingress LSRs and 1697 an egress LSRs through which a labeled packet traverses. The path of 1698 an explicit LSP is defined at the originating (ingress) node of the 1699 LSP. MPLS can use a signaling protocol such as RSVP or LDP to set up 1700 LSPs. 1702 MPLS is a very powerful technology for Internet traffic engineering 1703 because it supports explicit LSPs which allow constraint-based 1704 routing to be implemented efficiently in IP networks [AWD2]. The 1705 requirements for traffic engineering over MPLS are described in 1706 [RFC-2702]. Extensions to RSVP to support instantiation of explicit 1707 LSP are discussed in [AWD3]. Extensions to LDP, known as CR-LDP, to 1708 support explicit LSPs are presented in [JAM]. 1710 4.5.5 IP Performance Metrics 1712 The IETF IP Performance Metrics (IPPM) working group has been 1713 developing a set of standard metrics that can be used to monitor the 1714 quality, performance, and reliability of Internet services. These 1715 metrics can be applied by network operators, end-users, and 1716 independent testing groups to provide users and service providers 1717 with a common understanding of the performance and reliability of the 1718 Internet component 'clouds' they use/provide [RFC2330]. The criteria 1719 for performance metrics developed by the IPPM WG are described in 1720 [RFC2330]. Examples of performance metrics include one-way packet 1721 loss [RFC2680], one-way delay [RFC2679], and connectivity measures 1722 between two nodes [RFC2678]. Other metrics include second-order 1723 measures of packet loss and delay. 1725 Some of the performance metrics specified by the IPPM WG are useful 1726 for specifying Service Level Agreements (SLAs). SLAs are sets of 1727 service level objectives negotiated between users and service 1728 providers, wherein each objective is a combination of one or more 1729 performance metrics possibly subject to certain constraints. 1731 4.5.6 Flow Measurement 1733 The IETF Real Time Flow Measurement (RTFM) working group has produced 1734 an architecture document defining a method to specify traffic flows 1735 as well as a number of components for flow measurement (meters, meter 1736 readers, manager) [RFC-2722]. A flow measurement system enables 1737 network traffic flows to be measured and analyzed at the flow level 1738 for a variety of purposes. As noted in RFC-2722, a flow measurement 1739 system can be very useful in the following contexts: (1) 1740 understanding the behavior of existing networks, (2) planning for 1741 network development and expansion, (3) quantification of network 1742 performance, (4) verifying the quality of network service, and (5) 1743 attribution of network usage to users. 1745 A flow measurement system consists of meters, meter readers, and 1746 managers. A meter observe packets passing through a measurement 1747 point, classifies them into certain groups, accumulates certain usage 1748 data (such as the number of packets and bytes for each group), and 1749 stores the usage data in a flow table. A group may represent a user 1750 application, a host, a network, a group of networks, etc. A meter 1751 reader gathers usage data from various meters so it can be made 1752 available for analysis. A manager is responsible for configuring and 1753 controlling meters and meter readers. The instructions received by a 1754 meter from a manager include flow specification, meter control 1755 parameters, and sampling techniques. The instructions received by a 1756 meter reader from a manager include the address of the meter whose 1757 date is to be collected, the frequency of data collection, and the 1758 types of flows to be collected. 1760 4.5.7 Endpoint Congestion Management 1762 [RFC-3124] is intended to provide a set of congestion control 1763 mechanisms that transport protocols can use. It is also intended to 1764 develop mechanisms for unifying congestion control across a subset of 1765 an endpoint's active unicast connections (called a congestion group). 1766 A congestion manager continuously monitors the state of the path for 1767 each congestion group under its control. The manager uses that 1768 information to instruct a scheduler on how to partition bandwidth 1769 among the connections of that congestion group. 1771 4.6 Overview of ITU Activities Related to Traffic Engineering 1773 This section provides an overview of prior work within the ITU-T 1774 pertaining to traffic engineering in traditional telecommunications 1775 networks. 1777 ITU-T Recommendations E.600 [ITU-E600], E.701 [ITU-E701], and E.801 1778 [ITU-E801] address traffic engineering issues in traditional 1779 telecommunications networks. Recommendation E.600 provides a 1780 vocabulary for describing traffic engineering concepts, while E.701 1781 defines reference connections, Grade of Service (GOS), and traffic 1782 parameters for ISDN. Recommendation E.701 uses the concept of a 1783 reference connection to identify representative cases of different 1784 types of connections without describing the specifics of their actual 1785 realizations by different physical means. As defined in 1786 Recommendation E.600, "a connection is an association of resources 1787 providing means for communication between two or more devices in, or 1788 attached to, a telecommunication network." Also, E.600 defines "a 1789 resource as any set of physically or conceptually identifiable 1790 entities within a telecommunication network, the use of which can be 1791 unambiguously determined" [ITU-E600]. There can be different types 1792 of connections as the number and types of resources in a connection 1793 may vary. 1795 Typically, different network segments are involved in the path of a 1796 connection. For example, a connection may be local, national, or 1797 international. The purposes of reference connections are to clarify 1798 and specify traffic performance issues at various interfaces between 1799 different network domains. Each domain may consist of one or more 1800 service provider networks. 1802 Reference connections provide a basis to define grade of service 1803 (GoS) parameters related to traffic engineering within the ITU-T 1804 framework. As defined in E.600, "GoS refers to a number of traffic 1805 engineering variables which are used to provide a measure of the 1806 adequacy of a group of resources under specified conditions." These 1807 GoS variables may be probability of loss, dial tone, delay, etc. 1808 They are essential for network internal design and operation as well 1809 as for component performance specification. 1811 GoS is different from quality of service (QoS) in the ITU framework. 1812 QoS is the performance perceivable by a telecommunication service 1813 user and expresses the user's degree of satisfaction of the service. 1814 QoS parameters focus on performance aspects observable at the service 1815 access points and network interfaces, rather than their causes within 1816 the network. GoS, on the other hand, is a set of network oriented 1817 measures which characterize the adequacy of a group of resources 1818 under specified conditions. For a network to be effective in serving 1819 its users, the values of both GoS and QoS parameters must be related, 1820 with GoS parameters typically making a major contribution to the QoS. 1822 Recommendation E.600 stipulates that a set of GoS parameters must be 1823 selected and defined on an end-to-end basis for each major service 1824 category provided by a network to assist the network provider improve 1825 efficiency and effectiveness of the network. Based on a selected set 1826 of reference connections, suitable target values are assigned to the 1827 selected GoS parameters under normal and high load conditions. These 1828 end-to-end GoS target values are then apportioned to individual 1829 resource components of the reference connections for dimensioning 1830 purposes. 1832 4.7 Content Distribution 1834 The Internet is dominated by client-server interactions, especially 1835 Web traffic (in the future, more sophisticated media servers may 1836 become dominant). The location and performance of major information 1837 servers has a significant impact on the traffic patterns within the 1838 Internet as well as on the perception of service quality by end 1839 users. 1841 A number of dynamic load balancing techniques have been devised to 1842 improve the performance of replicated information servers. These 1843 techniques can cause spatial traffic characteristics to become more 1844 dynamic in the Internet because information servers can be 1845 dynamically picked based upon the location of the clients, the 1846 location of the servers, the relative utilization of the servers, the 1847 relative performance of different networks, and the relative 1848 performance of different parts of a network. This process of 1849 assignment of distributed servers to clients is called Traffic 1850 Directing. It functions at the application layer. 1852 Traffic Directing schemes that allocate servers in multiple 1853 geographically dispersed locations to clients may require empirical 1854 network performance statistics to make more effective decisions. In 1855 the future, network measurement systems may need to provide this type 1856 of information. The exact parameters needed are not yet defined. 1858 When congestion exists in the network, Traffic Directing and Traffic 1859 Engineering systems should act in a coordinated manner. This topic is 1860 for further study. 1862 The issues related to location and replication of information 1863 servers, particularly web servers, are important for Internet traffic 1864 engineering because these servers contribute a substantial proportion 1865 of Internet traffic. 1867 5.0 Taxonomy of Traffic Engineering Systems 1869 This section presents a short taxonomy of traffic engineering 1870 systems. A taxonomy of traffic engineering systems can be constructed 1871 based on traffic engineering styles and views as listed below: 1873 - Time-dependent vs State-dependent vs Event-dependent 1874 - Offline vs Online 1875 - Centralized vs Distributed 1876 - Local vs Global Information 1877 - Prescriptive vs Descriptive 1878 - Open Loop vs Closed Loop 1879 - Tactical vs Strategic 1881 These classification systems are described in greater detail in the 1882 following subsections of this document. 1884 5.1 Time-Dependent Versus State-Dependent Versus Event Dependent 1886 Traffic engineering methodologies can be classified as time-dependent 1887 or state-dependent or event-dependent. All TE schemes are considered 1888 to be dynamic in this document. Static TE implies that no traffic 1889 engineering methodology or algorithm is being applied. 1891 In the time-dependent TE, historical information based on periodic 1892 variations in traffic (such as time of day) is used to pre-program 1893 routing plans and other TE control mechanisms. Additionally, 1894 customer subscription or traffic projection may be used. Pre- 1895 programmed routing plans typically change on a relatively long time 1896 scale (e.g., diurnal). Time-dependent algorithms do not attempt to 1897 adapt to random variations in traffic or changing network conditions. 1898 An example of a time-dependent algorithm is a global centralized 1899 optimizer where the input to the system is a traffic matrix and 1900 multi-class QoS requirements as described [MR99]. 1902 State-dependent TE adapts the routing plans for packets based on the 1903 current state of the network. The current state of the network 1904 provides additional information on variations in actual traffic 1905 (i.e., perturbations from regular variations) that could not be 1906 predicted using historical information. Constraint-based routing is 1907 an example of state-dependent TE operating in a relatively long time 1908 scale. An example operating in a relatively short time scale is a 1909 load-balancing algorithm described in [MATE]. 1911 The state of the network can be based on parameters such as 1912 utilization, packet delay, packet loss, etc. These parameters can be 1913 obtained in several ways. For example, each router may flood these 1914 parameters periodically or by means of some kind of trigger to other 1915 routers. Another approach is for a particular router performing 1916 adaptive TE to send probe packets along a path to gather the state of 1917 that path. Still another approach is for a management system to 1918 gather relevant information from network elements. 1920 Expeditious and accurate gathering and distribution of state 1921 information is critical for adaptive TE due to the dynamic nature of 1922 network conditions. State-dependent algorithms may be applied to 1923 increase network efficiency and resilience. Time-dependent algorithms 1924 are more suitable for predictable traffic variations. On the other 1925 hand, state-dependent algorithms are more suitable for adapting to 1926 the prevailing network state. 1928 Event-dependent TE methods can also be used for TE path selection. 1929 Event-dependent TE methods are distinct from time-dependent and 1930 state-dependent TE methods in the manner in which paths are selected. 1931 These algorithms are adaptive and distributed in nature and typically 1932 use learning models to find good paths for TE in a network. While 1933 state-dependent TE models typically use available-link-bandwidth 1934 (ALB) flooding for TE path selection, event-dependent TE methods do 1935 not require ALB flooding. Rather, event-dependent TE methods 1936 typically search out capacity by learning models, as in the success- 1937 to-the-top (STT) method. ALB flooding can be resource intensive, 1938 since it requires link bandwidth to carry LSAs, processor capacity to 1939 process LSAs, and the overhead can limit area/autonomous system (AS) 1940 size. Modeling results suggest that event-dependent TE methods could 1941 lead to a reduction in ALB flooding overhead without loss of network 1942 throughput performance [ASH3]. 1944 5.2 Offline Versus Online 1946 Traffic engineering requires the computation of routing plans. The 1947 computation may be performed offline or online. The computation can 1948 be done offline for scenarios where routing plans need not be 1949 executed in real-time. For example, routing plans computed from 1950 forecast information may be computed offline. Typically, offline 1951 computation is also used to perform extensive searches on multi- 1952 dimensional solution spaces. 1954 Online computation is required when the routing plans must adapt to 1955 changing network conditions as in state-dependent algorithms. Unlike 1956 offline computation (which can be computationally demanding), online 1957 computation is geared toward relative simple and fast calculations to 1958 select routes, fine-tune the allocations of resources, and perform 1959 load balancing. 1961 5.3 Centralized Versus Distributed 1963 Centralized control has a central authority which determines routing 1964 plans and perhaps other TE control parameters on behalf of each 1965 router. The central authority collects the network-state information 1966 from all routers periodically and returns the routing information to 1967 the routers. The routing update cycle is a critical parameter 1968 directly impacting the performance of the network being controlled. 1969 Centralized control may need high processing power and high bandwidth 1970 control channels. 1972 Distributed control determines route selection by each router 1973 autonomously based on the routers view of the state of the network. 1974 The network state information may be obtained by the router using a 1975 probing method or distributed by other routers on a periodic basis 1976 using link state advertisements. Network state information may also 1977 be disseminated under exceptional conditions. 1979 5.4 Local Versus Global 1981 Traffic engineering algorithms may require local or global network- 1982 state information. 1984 Local information pertains to the state of a portion of the domain. 1985 Examples include the bandwidth and packet loss rate of a particular 1986 path. Local state information may be sufficient for certain 1987 instances of distributed-controlled TEs. 1989 Global information pertains to the state of the entire domain 1990 undergoing traffic engineering. Examples include a global traffic 1991 matrix and loading information on each link throughout the domain of 1992 interest. Global state information is typically required with 1993 centralized control. Distributed TE systems may also need global 1994 information in some cases. 1996 5.5 Prescriptive Versus Descriptive 1998 TE systems may also be classified as prescriptive or descriptive. 2000 Prescriptive traffic engineering evaluates alternatives and 2001 recommends a course of action. Prescriptive traffic engineering can 2002 be further categorized as either corrective or perfective. Corrective 2003 TE prescribes a course of action to address an existing or predicted 2004 anomaly. Perfective TE prescribes a course of action to evolve and 2005 improve network performance even when no anomalies are evident. 2007 Descriptive traffic engineering, on the other hand, characterizes the 2008 state of the network and assesses the impact of various policies 2009 without recommending any particular course of action. 2011 5.6 Open-Loop Versus Closed-Loop 2013 Open-loop traffic engineering control is where control action does 2014 not use feedback information from the current network state. The 2015 control action may use its own local information for accounting 2016 purposes, however. 2018 Closed-loop traffic engineering control is where control action 2019 utilizes feedback information from the network state. The feedback 2020 information may be in the form of historical information or current 2021 measurement. 2023 5.7 Tactical vs Strategic 2025 Tactical traffic engineering aims to address specific performance 2026 problems (such as hot-spots) that occur in the network from a 2027 tactical perspective, without consideration of overall strategic 2028 imperatives. Without proper planning and insights, tactical TE tends 2029 to be ad hoc in nature. 2031 Strategic traffic engineering approaches the TE problem from a more 2032 organized and systematic perspective, taking into consideration the 2033 immediate and longer term consequences of specific policies and 2034 actions. 2036 6.0 Recommendations for Internet Traffic Engineering 2038 This section describes high level recommendations for traffic 2039 engineering in the Internet. These recommendations are presented in 2040 general terms. 2042 The recommendations describe the capabilities needed to solve a 2043 traffic engineering problem or to achieve a traffic engineering 2044 objective. Broadly speaking, these recommendations can be categorized 2045 as either functional and non-functional recommendations. 2047 Functional recommendations for Internet traffic engineering describe 2048 the functions that a traffic engineering system should perform. These 2049 functions are needed to realize traffic engineering objectives by 2050 addressing traffic engineering problems. 2052 Non-functional recommendations for Internet traffic engineering 2053 relate to the quality attributes or state characteristics of a 2054 traffic engineering system. These recommendations may contain 2055 conflicting assertions and may sometimes be difficult to quantify 2056 precisely. 2058 6.1 Generic Non-functional Recommendations 2060 The generic non-functional recommendations for Internet traffic 2061 engineering include: usability, automation, scalability, stability, 2062 visibility, simplicity, efficiency, reliability, correctness, 2063 maintainability, extensibility, interoperability, and security. In a 2064 given context, some of these recommendations may be critical while 2065 others may be optional. Therefore, prioritization may be required 2066 during the development phase of a traffic engineering system (or 2067 components thereof) to tailor it to a specific operational context. 2069 In the following paragraphs, some of the aspects of the non- 2070 functional recommendations for Internet traffic engineering are 2071 summarized. 2073 Usability: Usability is a human factor aspect of traffic engineering 2074 systems. Usability refers to the ease with which a traffic 2075 engineering system can be deployed and operated. In general, it is 2076 desirable to have a TE system that can be readily deployed in an 2077 existing network. It is also desirable to have a TE system that is 2078 easy to operate and maintain. 2080 Automation: Whenever feasible, a traffic engineering system should 2081 automate as many traffic engineering functions as possible to 2082 minimize the amount of human effort needed to control and analyze 2083 operational networks. Automation is particularly imperative in large 2084 scale public networks because of the high cost of the human aspects 2085 of network operations and the high risk of network problems caused by 2086 human errors. Automation may entail the incorporation of automatic 2087 feedback and intelligence into some components of the traffic 2088 engineering system. 2090 Scalability: Contemporary public networks are growing very fast with 2091 respect to network size and traffic volume. Therefore, a TE system 2092 should be scalable to remain applicable as the network evolves. In 2093 particular, a TE system should remain functional as the network 2094 expands with regard to the number of routers and links, and with 2095 respect to the traffic volume. A TE system should have a scalable 2096 architecture, should not adversely impair other functions and 2097 processes in a network element, and should not consume too much 2098 network resources when collecting and distributing state information 2099 or when exerting control. 2101 Stability: Stability is a very important consideration in traffic 2102 engineering systems that respond to changes in the state of the 2103 network. State-dependent traffic engineering methodologies typically 2104 mandate a tradeoff between responsiveness and stability. It is 2105 strongly recommended that when tradeoffs are warranted between 2106 responsiveness and stability, that the tradeoff should be made in 2107 favor of stability (especially in public IP backbone networks). 2109 Flexibility: A TE system should be flexible to allow for changes in 2110 optimization policy. In particular, a TE system should provide 2111 sufficient configuration options so that a network administrator can 2112 tailor the TE system to a particular environment. It may also be 2113 desirable to have both online and offline TE subsystems which can be 2114 independently enabled and disabled. TE systems that are used in 2115 multi-class networks should also have options to support class based 2116 performance evaluation and optimization. 2118 Visibility: As part of the TE system, mechanisms should exist to 2119 collect statistics from the network and to analyze these statistics 2120 to determine how well the network is functioning. Derived statistics 2121 such as traffic matrices, link utilization, latency, packet loss, and 2122 other performance measures of interest which are determined from 2123 network measurements can be used as indicators of prevailing network 2124 conditions. Other examples of status information which should be 2125 observed include existing functional routing information 2126 (additionally, in the context of MPLS existing LSP routes), etc. 2128 Simplicity: Generally, a TE system should be as simple as possible. 2129 More importantly, the TE system should be relatively easy to use 2130 (i.e., clean, convenient, and intuitive user interfaces). Simplicity 2131 in user interface does not necessarily imply that the TE system will 2132 use naive algorithms. When complex algorithms and internal structures 2133 are used, such complexities should be hidden as much as possible from 2134 the network administrator through the user interface. 2136 Interoperability: Whenever feasible, traffic engineering systems and 2137 their components should be developed with open standards based 2138 interfaces to allow interoperation with other systems and components. 2140 Security: Security is a critical consideration in traffic engineering 2141 systems. Such traffic engineering systems typically exert control 2142 over certain functional aspects of the network to achieve the desired 2143 performance objectives. Therefore, adequate measures must be taken to 2144 safeguard the integrity of the traffic engineering system. Adequate 2145 measures must also be taken to protect the network from 2146 vulnerabilities that originate from security breaches and other 2147 impairments within the traffic engineering system. 2149 The remainder of this section will focus on some of the high level 2150 functional recommendations for traffic engineering. 2152 6.2 Routing Recommendations 2154 Routing control is a significant aspect of Internet traffic 2155 engineering. Routing impacts many of the key performance measures 2156 associated with networks, such as throughput, delay, and utilization. 2157 Generally, it is very difficult to provide good service quality in a 2158 wide area network without effective routing control. A desirable 2159 routing system is one that takes traffic characteristics and network 2160 constraints into account during route selection while maintaining 2161 stability. 2163 Traditional shortest path first (SPF) interior gateway protocols are 2164 based on shortest path algorithms and have limited control 2165 capabilities for traffic engineering [RFC-2702, AWD2]. These 2166 limitations include : 2168 1. The well known issues with pure SPF protocols, which 2169 do not take network constraints and traffic characteristics 2170 into account during route selection. For example, since IGPs 2171 always use the shortest paths (based on administratively 2172 assigned link metrics) to forward traffic, load sharing cannot 2173 be accomplished among paths of different costs. Using shortest 2174 paths to forward traffic conserves network resources, but may 2175 cause the following problems: 1) If traffic from a source to a 2176 destination exceeds the capacity of a link along the shortest 2177 path, the link (hence the shortest path) becomes congested while 2178 a longer path between these two nodes may be under-utilized; 2179 2) the shortest paths from different sources can overlap at some 2180 links. If the total traffic from the sources exceeds the 2181 capacity of any of these links, congestion will occur. Problems 2182 can also occur because traffic demand changes over time but 2183 network topology and routing configuration cannot be changed as 2184 rapidly. This causes the network topology and routing 2185 configuration to become sub-optimal over time, which may result 2186 in persistent congestion problems. 2188 2. The Equal-Cost Multi-Path (ECMP) capability of SPF IGPs supports 2189 sharing of traffic among equal cost paths between two nodes. 2190 However, ECMP attempts to divide the traffic as equally as 2191 possible among the equal cost shortest paths. Generally, ECMP 2192 does not support configurable load sharing ratios among equal 2193 cost paths. The result is that one of the paths may carry 2194 significantly more traffic than other paths because it 2195 may also carry traffic from other sources. This situation can 2196 result in congestion along the path that carries more traffic. 2198 3. Modifying IGP metrics to control traffic routing tends to 2199 have network-wide effect. Consequently, undesirable and 2200 unanticipated traffic shifts can be triggered as a result. 2202 Because of these limitations, new capabilities are needed to enhance 2203 the routing function in IP networks. Some of these capabilities have 2204 been described elsewhere and are summarized below. 2206 Constraint-based routing is desirable to evolve the routing 2207 architecture of IP networks, especially public IP backbones with 2208 complex topologies [RFC-2702]. Constraint-based routing computes 2209 routes to fulfill requirements subject to constraints. Constraints 2210 may include bandwidth, hop count, delay, and administrative policy 2211 instruments such as resource class attributes [RFC-2702, RFC-2386]. 2212 This makes it possible to select routes that satisfy a given set of 2213 requirements subject to network and administrative policy 2214 constraints. Routes computed through constraint-based routing are not 2215 necessarily the shortest paths. Constraint-based routing works best 2216 with path oriented technologies that support explicit routing, such 2217 as MPLS. 2219 Constraint-based routing can also be used as a way to redistribute 2220 traffic onto the infrastructure (even for best effort traffic). For 2221 example, if the bandwidth requirements for path selection and 2222 reservable bandwidth attributes of network links are appropriately 2223 defined and configured, then congestion problems caused by uneven 2224 traffic distribution may be avoided or reduced. In this way, the 2225 performance and efficiency of the network can be improved. 2227 A number of enhancements are needed to conventional link state IGPs, 2228 such as OSPF and IS-IS, to allow them to distribute additional state 2229 information required for constraint-based routing. These extensions 2230 to OSPF were described in [KATZ] and to IS-IS in [SMIT]. 2231 Essentially, these enhancements require the propagation of additional 2232 information in link state advertisements. Specifically, in addition 2233 to normal link-state information, an enhanced IGP is required to 2234 propagate topology state information needed for constraint-based 2235 routing. Some of the additional topology state information include 2236 link attributes such as reservable bandwidth and link resource class 2237 attribute (an administratively specified property of the link). The 2238 resource class attribute concept was defined in [RFC-2702]. The 2239 additional topology state information is carried in new TLVs and 2240 sub-TLVs in IS-IS, or in the Opaque LSA in OSPF [SMIT, KATZ]. 2242 An enhanced link-state IGP may flood information more frequently than 2243 a normal IGP. This is because even without changes in topology, 2244 changes in reservable bandwidth or link affinity can trigger the 2245 enhanced IGP to initiate flooding. A tradeoff is typically required 2246 between the timeliness of the information flooded and the flooding 2247 frequency to avoid excessive consumption of link bandwidth and 2248 computational resources, and more importantly, to avoid instability. 2250 In a TE system, it is also desirable for the routing subsystem to 2251 make the load splitting ratio among multiple paths (with equal cost 2252 or different cost) configurable. This capability gives network 2253 administrators more flexibility in the control of traffic 2254 distribution across the network. It can be very useful for 2255 avoiding/relieving congestion in certain situations. Examples can be 2256 found in [XIAO]. 2258 The routing system should also have the capability to control the 2259 routes of subsets of traffic without affecting the routes of other 2260 traffic if sufficient resources exist for this purpose. This 2261 capability allows a more refined control over the distribution of 2262 traffic across the network. For example, the ability to move traffic 2263 from a source to a destination away from its original path to another 2264 path (without affecting other traffic paths) allows traffic to be 2265 moved from resource-poor network segments to resource-rich segments. 2266 Path oriented technologies such as MPLS inherently support this 2267 capability as discussed in [AWD2]. 2269 Additionally, the routing subsystem should be able to select 2270 different paths for different classes of traffic (or for different 2271 traffic behavior aggregates) if the network supports multiple classes 2272 of service (different behavior aggregates). 2274 6.3 Traffic Mapping Recommendations 2276 Traffic mapping pertains to the assignment of traffic workload onto 2277 pre-established paths to meet certain requirements. Thus, while 2278 constraint-based routing deals with path selection, traffic mapping 2279 deals with the assignment of traffic to established paths which may 2280 have been selected by constraint-based routing or by some other 2281 means. Traffic mapping can be performed by time-dependent or state- 2282 dependent mechanisms, as described in Section 5.1. 2284 An important aspect of the traffic mapping function is the ability to 2285 establish multiple paths between an originating node and a 2286 destination node, and the capability to distribute the traffic 2287 between the two nodes across the paths according to some policies. A 2288 pre-condition for this scheme is the existence of flexible mechanisms 2289 to partition traffic and then assign the traffic partitions onto the 2290 parallel paths. This requirement was noted in [RFC-2702]. When 2291 traffic is assigned to multiple parallel paths, it is recommended 2292 that special care should be taken to ensure proper ordering of 2293 packets belonging to the same application (or micro-flow) at the 2294 destination node of the parallel paths. 2296 As a general rule, mechanisms that perform the traffic mapping 2297 functions should aim to map the traffic onto the network 2298 infrastructure to minimize congestion. If the total traffic load 2299 cannot be accommodated, or if the routing and mapping functions 2300 cannot react fast enough to changing traffic conditions, then a 2301 traffic mapping system may rely on short time scale congestion 2302 control mechanisms (such as queue management, scheduling, etc) to 2303 mitigate congestion. Thus, mechanisms that perform the traffic 2304 mapping functions should complement existing congestion control 2305 mechanisms. In an operational network, it is generally desirable to 2306 map the traffic onto the infrastructure such that intra-class and 2307 inter-class resource contention are minimized. 2309 When traffic mapping techniques that depend on dynamic state feedback 2310 (e.g. MATE and such like) are used, special care must be taken to 2311 guarantee network stability. 2313 6.4 Measurement Recommendations 2315 The importance of measurement in traffic engineering has been 2316 discussed throughout this document. Mechanisms should be provided to 2317 measure and collect statistics from the network to support the 2318 traffic engineering function. Additional capabilities may be needed 2319 to help in the analysis of the statistics. The actions of these 2320 mechanisms should not adversely affect the accuracy and integrity of 2321 the statistics collected. The mechanisms for statistical data 2322 acquisition should also be able to scale as the network evolves. 2324 Traffic statistics may be classified according to long-term or 2325 short-term time scales. Long-term time scale traffic statistics are 2326 very useful for traffic engineering. Long-term time scale traffic 2327 statistics may capture or reflect periodicity in network workload 2328 (such as hourly, daily, and weekly variations in traffic profiles) as 2329 well as traffic trends. Aspects of the monitored traffic statistics 2330 may also depict class of service characteristics for a network 2331 supporting multiple classes of service. Analysis of the long-term 2332 traffic statistics MAY yield secondary statistics such as busy hour 2333 characteristics, traffic growth patterns, persistent congestion 2334 problems, hot-spot, and imbalances in link utilization caused by 2335 routing anomalies. 2337 A mechanism for constructing traffic matrices for both long-term and 2338 short-term traffic statistics should be in place. In multi-service IP 2339 networks, the traffic matrices may be constructed for different 2340 service classes. Each element of a traffic matrix represents a 2341 statistic of traffic flow between a pair of abstract nodes. An 2342 abstract node may represent a router, a collection of routers, or a 2343 site in a VPN. 2345 Measured traffic statistics should provide reasonable and reliable 2346 indicators of the current state of the network on the short-term 2347 scale. Some short term traffic statistics may reflect link 2348 utilization and link congestion status. Examples of congestion 2349 indicators include excessive packet delay, packet loss, and high 2350 resource utilization. Examples of mechanisms for distributing this 2351 kind of information include SNMP, probing techniques, FTP, IGP link 2352 state advertisements, etc. 2354 6.5 Network Survivability 2356 Network survivability refers to the capability of a network to 2357 maintain service continuity in the presence of faults. This can be 2358 accomplished by promptly recovering from network impairments and 2359 maintaining the required QoS for existing services after recovery. 2360 Survivability has become an issue of great concern within the 2361 Internet community due to the increasing demands to carry mission 2362 critical traffic, real-time traffic, and other high priority traffic 2363 over the Internet. Survivability can be addressed at the device level 2364 by developing network elements that are more reliable; and at the 2365 network level by incorporating redundancy into the architecture, 2366 design, and operation of networks. It is recommended that a 2367 philosophy of robustness and survivability should be adopted in the 2368 architecture, design, and operation of traffic engineering that 2369 control IP networks (especially public IP networks). Because 2370 different contexts may demand different levels of survivability, the 2371 mechanisms developed to support network survivability should be 2372 flexible so that they can be tailored to different needs. 2374 Failure protection and restoration capabilities have become available 2375 from multiple layers as network technologies have continued to 2376 improve. At the bottom of the layered stack, optical networks are now 2377 capable of providing dynamic ring and mesh restoration functionality 2378 at the wavelength level as well as traditional protection 2379 functionality. At the SONET/SDH layer survivability capability is 2380 provided with Automatic Protection Switching (APS) as well as self- 2381 healing ring and mesh architectures. Similar functionality is 2382 provided by layer 2 technologies such as ATM (generally with slower 2383 mean restoration times). Rerouting is traditionally used at the IP 2384 layer to restore service following link and node outages. Rerouting 2385 at the IP layer occurs after a period of routing convergence which 2386 may require seconds to minutes to complete. Some new developments in 2387 the MPLS context make it possible to achieve recovery at the IP layer 2388 prior to convergence [SHAR]. 2390 To support advanced survivability requirements, path-oriented 2391 technologies such a MPLS can be used to enhance the survivability of 2392 IP networks in a potentially cost effective manner. The advantages of 2393 path oriented technologies such as MPLS for IP restoration becomes 2394 even more evident when class based protection and restoration 2395 capabilities are required. 2397 Recently, a common suite of control plane protocols has been proposed 2398 for both MPLS and optical transport networks under the acronym 2399 Multi-protocol Lambda Switching [AWD1]. This new paradigm of Multi- 2400 protocol Lambda Switching will support even more sophisticated mesh 2401 restoration capabilities at the optical layer for the emerging IP 2402 over WDM network architectures. 2404 Another important aspect regarding multi-layer survivability is that 2405 technologies at different layers provide protection and restoration 2406 capabilities at different temporal granularities (in terms of time 2407 scales) and at different bandwidth granularity (from packet-level to 2408 wavelength level). Protection and restoration capabilities can also 2409 be sensitive to different service classes and different network 2410 utility models. 2412 The impact of service outages varies significantly for different 2413 service classes depending upon the effective duration of the outage. 2414 The duration of an outage can vary from milliseconds (with minor 2415 service impact) to seconds (with possible call drops for IP telephony 2416 and session time-outs for connection oriented transactions) to 2417 minutes and hours (with potentially considerable social and business 2418 impact). 2420 Coordinating different protection and restoration capabilities across 2421 multiple layers in a cohesive manner to ensure network survivability 2422 is maintained at reasonable cost is a challenging task. Protection 2423 and restoration coordination across layers may not always be 2424 feasible, because networks at different layers may belong to 2425 different administrative domains. 2427 The following paragraphs present some of the general recommendations 2428 for protection and restoration coordination. 2430 - Protection and restoration capabilities from different layers 2431 should be coordinated whenever feasible and appropriate to 2432 provide network survivability in a flexible and cost effective 2433 manner. Minimization of function duplication across layers is 2434 one way to achieve the coordination. Escalation of alarms and 2435 other fault indicators from lower to higher layers may also 2436 be performed in a coordinated manner. A temporal order of 2437 restoration trigger timing at different layers is another way 2438 to coordinate multi-layer protection/restoration. 2440 - Spare capacity at higher layers is often regarded as working 2441 traffic at lower layers. Placing protection/restoration 2442 functions in many layers may increase redundancy and robustness, 2443 but it should not result in significant and avoidable 2444 inefficiencies in network resource utilization. 2446 - It is generally desirable to have protection and restoration 2447 schemes that are bandwidth efficient. 2449 - Failure notification throughout the network should be timely 2450 and reliable. 2452 - Alarms and other fault monitoring and reporting capabilities 2453 should be provided at appropriate layers. 2455 6.5.1 Survivability in MPLS Based Networks 2457 MPLS is an important emerging technology that enhances IP networks in 2458 terms of features, capabilities, and services. Because MPLS is path- 2459 oriented it can potentially provide faster and more predictable 2460 protection and restoration capabilities than conventional hop by hop 2461 routed IP systems. This subsection describes some of the basic 2462 aspects and recommendations for MPLS networks regarding protection 2463 and restoration. See [SHAR] for a more comprehensive discussion on 2464 MPLS based recovery. 2466 Protection types for MPLS networks can be categorized as link 2467 protection, node protection, path protection, and segment protection. 2469 - Link Protection: The objective for link protection is to protect 2470 an LSP from a given link failure. Under link protection, the path 2471 of the protection or backup LSP (the secondary LSP) is disjoint 2472 from the path of the working or operational LSP at the particular 2473 link over which protection is required. When the protected link 2474 fails, traffic on the working LSP is switched over to the 2475 protection LSP at the head-end of the failed link. This is a local 2476 repair method which can be fast. It might be more appropriate in 2477 situations where some network elements along a given path are 2478 less reliable than others. 2480 - Node Protection: The objective of LSP node protection is to protect 2481 an LSP from a given node failure. Under node protection, the path 2482 of the protection LSP is disjoint from the path of the working LSP 2483 at the particular node to be protected. The secondary path is 2484 also disjoint from the primary path at all links associated with 2485 the node to be protected. When the node fails, traffic on the 2486 working LSP is switched over to the protection LSP at the upstream 2487 LSR directly connected to the failed node. 2489 - Path Protection: The goal of LSP path protection is to protect an 2490 LSP from failure at any point along its routed path. Under path 2491 protection, the path of the protection LSP is completely disjoint 2492 from the path of the working LSP. The advantage of path protection 2493 is that the backup LSP protects the working LSP from all possible 2494 link and node failures along the path, except for failures that 2495 might occur at the ingress and egress LSRs, or for correlated 2496 failures that might impact both working and backup paths 2497 simultaneously. Additionally, since the path selection is 2498 end-to-end, path protection might be more efficient in terms of 2499 resource usage than link or node protection. However, path 2500 protection may be slower than link and node protection in general. 2502 - Segment Protection: An MPLS domain may be partitioned into multiple 2503 protection domains whereby a failure in a protection domain is 2504 rectified within that domain. In cases where an LSP traverses 2505 multiple protection domains, a protection mechanism within a domain 2506 only needs to protect the segment of the LSP that lies within the 2507 domain. Segment protection will generally be faster than path 2508 protection because recovery generally occurs closer to the fault. 2510 6.5.2 Protection Option 2512 Another issue to consider is the concept of protection options. The 2513 protection option uses the notation m:n protection where m is the 2514 number of protection LSPs used to protect n working LSPs. Feasible 2515 protection options follow. 2517 - 1:1: one working LSP is protected/restored by one protection LSP. 2519 - 1:n: one protection LSP is used to protect/restore n working LSPs. 2521 - n:1: one working LSP is protected/restored by n protection LSPs, 2522 possibly with configurable load splitting ratio. When more than 2523 one protection LSP is used, it may be desirable to share the 2524 traffic across the protection LSPs when the working LSP fails to 2525 satisfy the bandwidth requirement of the traffic trunk associated 2526 with the working LSP. This may be especially useful when it is 2527 not feasible to find one path that can satisfy the bandwidth 2528 requirement of the primary LSP. 2530 - 1+1: traffic is sent concurrently on both the working LSP and the 2531 protection LSP. In this case, the egress LSR selects one of the two 2532 LSPs based on a local traffic integrity decision process, which 2533 compares the traffic received from both the working and the 2534 protection LSP and identifies discrepancies. It is unlikely that 2535 this option would be used extensively in IP networks due to its 2536 resource utilization inefficiency. However, if bandwidth becomes 2537 plentiful and cheap, then this option might become quite viable and 2538 attractive in IP networks. 2540 6.6 Traffic Engineering in Diffserv Environments 2542 This section provides an overview of the traffic engineering features 2543 and recommendations that are specifically pertinent to Differentiated 2544 Services (Diffserv) [RFC-2475] capable IP networks. 2546 Increasing requirements to support multiple classes of traffic, such 2547 as best effort and mission critical data, in the Internet calls for 2548 IP networks to differentiate traffic according to some criteria, and 2549 to accord preferential treatment to certain types of traffic. Large 2550 numbers of flows can be aggregated into a few behavior aggregates 2551 based on some criteria in terms of common performance requirements in 2552 terms of packet loss ratio, delay, and jitter; or in terms of common 2553 fields within the IP packet headers. 2555 As Diffserv evolves and becomes deployed in operational networks, 2556 traffic engineering will be critical to ensuring that SLAs defined 2557 within a given Diffserv service model are met. Classes of service 2558 (CoS) can be supported in a Diffserv environment by concatenating 2559 per-hop behaviors (PHBs) along the routing path, using service 2560 provisioning mechanisms, and by appropriately configuring edge 2561 functionality such as traffic classification, marking, policing, and 2562 shaping. PHB is the forwarding behavior that a packet receives at a 2563 DS node (a Diffserv-compliant node). This is accomplished by means of 2564 buffer management and packet scheduling mechanisms. In this context, 2565 packets belonging to a class are those that are members of a 2566 corresponding ordering aggregate. 2568 Traffic engineering can be used as a compliment to Diffserv 2569 mechanisms to improve utilization of network resources, but not as a 2570 necessary element in general. When traffic engineering is used, it 2571 can be operated on an aggregated basis across all service classes 2572 [MPLS-DIFF] or on a per service class basis. The former is used to 2573 provide better distribution of the aggregate traffic load over the 2574 network resources. (See [MPLS_DIFF] for detailed mechanisms to 2575 support aggregate traffic engineering.) The latter case is discussed 2576 below since it is specific to the Diffserv environment, with so 2577 called Diffserv-aware traffic engineering [DIFF_TE]. 2579 For some Diffserv networks, it may be desirable to control the 2580 performance of some service classes by enforcing certain 2581 relationships between the traffic workload contributed by each 2582 service class and the amount of network resources allocated or 2583 provisioned for that service class. Such relationships between 2584 demand and resource allocation can be enforced using a combination 2585 of, for example: (1) traffic engineering mechanisms on a per service 2586 class basis that enforce the desired relationship between the amount 2587 of traffic contributed by a given service class and the resources 2588 allocated to that class and (2) mechanisms that dynamically adjust 2589 the resources allocated to a given service class to relate to the 2590 amount of traffic contributed by that service class. 2592 It may also be desirable to limit the performance impact of high 2593 priority traffic on relatively low priority traffic. This can be 2594 achieved by, for example, controlling the percentage of high priority 2595 traffic that is routed through a given link. Another way to 2596 accomplish this is to increase link capacities appropriately so that 2597 lower priority traffic can still enjoy adequate service quality. When 2598 the ratio of traffic workload contributed by different service 2599 classes vary significantly from router to router, it may not suffice 2600 to rely exclusively on conventional IGP routing protocols or on 2601 traffic engineering mechanisms that are insensitive to different 2602 service classes. Instead, it may be desirable to perform traffic 2603 engineering, especially routing control and mapping functions, on a 2604 per service class basis. One way to accomplish this in a domain that 2605 supports both MPLS and Diffserv is to define class specific LSPs and 2606 to map traffic from each class onto one or more LSPs that correspond 2607 to that service class. An LSP corresponding to a given service class 2608 can then be routed and protected/restored in a class dependent 2609 manner, according to specific policies. 2611 Performing traffic engineering on a per class basis may require 2612 certain per-class parameters to be distributed. Note that it is 2613 common to have some classes to share some aggregate constraint (e.g. 2614 maximum bandwidth requirement) without enforcing the constraint on 2615 each individual class. These classes then can be grouped into a 2616 class-type and per-class-type parameters can be distributed instead 2617 to improve scalability. It also allows better bandwidth sharing 2618 between classes in the same class-type. A class-type is a set of 2619 classes that satisfy the following two conditions: 2621 1) Classes in the same class-type have common aggregate requirements 2622 to satisfy required performance levels. 2624 2) There is no requirement to be enforced at the level of individual 2625 class in the class-type. Note that it is still possible, 2626 nevertheless, to implement some priority policies for classes in the 2627 same class-type to permit preferential access to the class-type 2628 bandwidth through the use of preemption priorities. 2630 An example of the class-type can be a low-loss class-type that 2631 includes both AF1-based and AF2-based Ordering Aggregates. With such 2632 a class-type, one may implement some priority policy which assigns 2633 higher preemption priority to AF1-based traffic trunks over AF2-based 2634 ones, vice versa, or the same priority. 2636 See [DIFF-TE] for detailed requirements on Diffserv-aware traffic 2637 engineering. 2639 6.7 Network Controllability 2641 Off-line (and on-line) traffic engineering considerations would be of 2642 limited utility if the network could not be controlled effectively to 2643 implement the results of TE decisions and to achieve desired network 2644 performance objectives. Capacity augmentation is a coarse grained 2645 solution to traffic engineering issues. However, it is simple and may 2646 be advantageous if bandwidth is abundant and cheap or if the current 2647 or expected network workload demands it. However, bandwidth is not 2648 always abundant and cheap, and the workload may not always demand 2649 additional capacity. Adjustments of administrative weights and other 2650 parameters associated with routing protocols provide finer grained 2651 control, but is difficult to use and imprecise because of the routing 2652 interactions that occur across the network. In certain network 2653 contexts, more flexible, finer grained approaches which provide more 2654 precise control over the mapping of traffic to routes and over the 2655 selection and placement of routes may be appropriate and useful. 2657 Control mechanisms can be manual (e.g. administrative configuration), 2658 partially-automated (e.g. scripts) or fully-automated (e.g. policy 2659 based management systems). Automated mechanisms are particularly 2660 required in large scale networks. Multi-vendor interoperability can 2661 be facilitated by developing and deploying standardized management 2662 systems (e.g. standard MIBs) and policies (PIBs) to support the 2663 control functions required to address traffic engineering objectives 2664 such as load distribution and protection/restoration. 2666 Network control functions should be secure, reliable, and stable as 2667 these are often needed to operate correctly in times of network 2668 impairments (e.g. during network congestion or security attacks). 2670 7.0 Inter-Domain Considerations 2672 Inter-domain traffic engineering is concerned with the performance 2673 optimization for traffic that originates in one administrative domain 2674 and terminates in a different one. 2676 Traffic exchange between autonomous systems in the Internet occurs 2677 through exterior gateway protocols. Currently, BGP-4 [BGP4] is the 2678 standard exterior gateway protocol for the Internet. BGP-4 provides 2679 a number of attributes (e.g. local preference, AS path, and MED) and 2680 capabilities (e.g. route filtering) that can be used for inter-domain 2681 traffic engineering. These mechanisms are generally effective, but 2682 they are usually applied in a trial-and-error fashion. A systematic 2683 approach for inter-domain traffic engineering is yet to be devised. 2685 Inter-domain traffic engineering is inherently more difficult than 2686 intra-domain TE under the current Internet architecture. The reasons 2687 for this are both technical and administrative. Technically, while 2688 topology and link state information are helpful for mapping traffic 2689 more effectively, BGP does not propagate such information across 2690 domain boundaries for stability and scalability reasons. 2691 Administratively, there are differences in operating costs and 2692 network capacities between domains. Generally, what may be considered 2693 a good solution in one domain may not necessarily be a good solution 2694 in another domain. Moreover, it would generally be considered 2695 inadvisable for one domain to permit another domain to influence the 2696 routing and management of traffic in its network. 2698 MPLS TE-tunnels (explicit LSPs) can potentially add a degree of 2699 flexibility in the selection of exit points for inter-domain routing. 2700 The concept of relative and absolute metrics can be applied to this 2701 purpose. The idea is that if BGP attributes are defined such that the 2702 BGP decision process depends on IGP metrics to select exit points for 2703 inter-domain traffic, then some inter-domain traffic destined to a 2704 given peer network can be made to prefer a specific exit point by 2705 establishing a TE-tunnel between the router making the selection to 2706 the peering point via a TE-tunnel and assigning the TE-tunnel a 2707 metric which is smaller than the IGP cost to all other peering 2708 points. If a peer accepts and processes MEDs, then a similar MPLS 2709 TE-tunnel based scheme can be applied to cause certain entrance 2710 points to be preferred by setting MED to be an IGP cost, which has 2711 been modified by the tunnel metric. 2713 Similar to intra-domain TE, inter-domain TE is best accomplished when 2714 a traffic matrix can be derived to depict the volume of traffic from 2715 one autonomous system to another. 2717 Generally, redistribution of inter-domain traffic requires 2718 coordination between peering partners. An export policy in one domain 2719 that results in load redistribution across peer points with another 2720 domain can significantly affect the local traffic matrix inside the 2721 domain of the peering partner. This, in turn, will affect the intra- 2722 domain TE due to changes in the spatial distribution traffic. 2723 Therefore, it is mutually beneficial for peering partners to 2724 coordinate with each other before attempting any policy changes that 2725 may result in significant shifts in inter-domain traffic. In certain 2726 contexts, this coordination can be quite challenging due to technical 2727 and non- technical reasons. 2729 It is a matter of speculation as to whether MPLS, or similar 2730 technologies, can be extended to allow selection of constrained paths 2731 across domain boundaries. 2733 8.0 Overview of Contemporary TE Practices in Operational IP Networks 2735 This section provides an overview of some contemporary traffic 2736 engineering practices in IP networks. The focus is primarily on the 2737 aspects that pertain to the control of the routing function in 2738 operational contexts. The intent here is to provide an overview of 2739 the commonly used practices. The discussion is not intended to be 2740 exhaustive. 2742 Currently, service providers apply many of the traffic engineering 2743 mechanisms discussed in this document to optimize the performance of 2744 their IP networks. These techniques include capacity planning for 2745 long time scales, routing control using IGP metrics and MPLS for 2746 medium time scales, the overlay model also for medium time scales, 2747 and traffic management mechanisms for short time scale. 2749 When a service provider plans to build an IP network, or expand the 2750 capacity of an existing network, effective capacity planning should 2751 be an important component of the process. Such plans may take the 2752 following aspects into account: location of new nodes if any, 2753 existing and predicted traffic patterns, costs, link capacity, 2754 topology, routing design, and survivability. 2756 Performance optimization of operational networks is usually an 2757 ongoing process in which traffic statistics, performance parameters, 2758 and fault indicators are continually collected from the network. 2759 These empirical data are then analyzed and used to trigger various 2760 traffic engineering mechanisms. For example, IGP parameters, e.g., 2761 OSPF or IS-IS metrics, can be adjusted based on manual computations 2762 or based on the output of some traffic engineering support tools. 2763 Such tools may use the following as input the: traffic matrix, 2764 network topology, and network performance objective(s). Tools that 2765 perform what-if analysis can also be used to assist the TE process by 2766 allowing various scenarios to be reviewed before a new set of 2767 configurations are implemented in the operational network. 2769 The overlay model (IP over ATM or IP over Frame relay) is another 2770 approach which is commonly used in practice [AWD2]. The IP over ATM 2771 technique is no longer viewed favorably due to recent advances in 2772 MPLS and router hardware technology. 2774 Deployment of MPLS for traffic engineering applications has commenced 2775 in some service provider networks. One operational scenario is to 2776 deploy MPLS in conjunction with an IGP (IS-IS-TE or OSPF-TE) that 2777 supports the traffic engineering extensions, in conjunction with 2778 constraint-based routing for explicit route computations, and a 2779 signaling protocol (e.g. RSVP-TE or CRLDP) for LSP instantiation. 2781 In contemporary MPLS traffic engineering contexts, network 2782 administrators specify and configure link attributes and resource 2783 constraints such as maximum reservable bandwidth and resource class 2784 attributes for links (interfaces) within the MPLS domain. A link 2785 state protocol that supports TE extensions (IS-IS-TE or OSPF-TE) is 2786 used to propagate information about network topology and link 2787 attribute to all routers in the routing area. Network administrators 2788 also specify all the LSPs that are to originate each router. For each 2789 LSP, the network administrator specifies the destination node and the 2790 attributes of the LSP which indicate the requirements that to be 2791 satisfied during the path selection process. Each router then uses a 2792 local constraint-based routing process to compute explicit paths for 2793 all LSPs originating from it. Subsequently, a signaling protocol is 2794 used to instantiate the LSPs. By assigning proper bandwidth values to 2795 links and LSPs, congestion caused by uneven traffic distribution can 2796 generally be avoided or mitigated. 2798 The bandwidth attributes of LSPs used for traffic engineering can be 2799 updated periodically. The basic concept is that the bandwidth 2800 assigned to an LSP should relate in some manner to the bandwidth 2801 requirements of traffic that actually flows through the LSP. The 2802 traffic attribute of an LSP can be modified to accommodate traffic 2803 growth and persistent traffic shifts. If network congestion occurs 2804 due to some unexpected events, existing LSPs can be rerouted to 2805 alleviate the situation or network administrator can configure new 2806 LSPs to divert some traffic to alternative paths. The reservable 2807 bandwidth of the congested links can also be reduced to force some 2808 LSPs to be rerouted to other paths. 2810 In an MPLS domain, a traffic matrix can also be estimated by 2811 monitoring the traffic on LSPs. Such traffic statistics can be used 2812 for a variety of purposes including network planning and network 2813 optimization. Current practice suggests that deploying an MPLS 2814 network consisting of hundreds of routers and thousands of LSPs is 2815 feasible. In summary, recent deployment experience suggests that MPLS 2816 approach is very effective for traffic engineering in IP networks 2817 [XIAO]. 2819 9.0 Conclusion 2821 This document described principles for traffic engineering in the 2822 Internet. It presented an overview of some of the basic issues 2823 surrounding traffic engineering in IP networks. The context of TE was 2824 described, a TE process models and a taxonomy of TE styles were 2825 presented. A brief historical review of pertinent developments 2826 related to traffic engineering was provided. A survey of contemporary 2827 TE techniques in operational networks was presented. Additionally, 2828 the document specified a set of generic requirements, 2829 recommendations, and options for Internet traffic engineering. 2831 10.0 Security Considerations 2833 This document does not introduce new security issues. 2835 11.0 Acknowledgments 2837 The authors would like to thank Jim Boyle for inputs on the 2838 recommendations section, Francois Le Faucheur for inputs on Diffserv 2839 aspects, Blaine Christian for inputs on measurement, Gerald Ash for 2840 inputs on routing in telephone networks and for text on event- 2841 dependent TE methods , and Steven Wright for inputs on network 2842 controllability. Special thanks to Randy Bush for proposing the TE 2843 taxonomy based on "tactical vs strategic" methods. The subsection 2844 describing an "Overview of ITU Activities Related to Traffic 2845 Engineering" was adapted from a contribution by Waisum Lai. Useful 2846 feedback and pointers to relevant materials were provided by J. Noel 2847 Chiappa. Additional comments were provided by Glenn Grotefeld during 2848 the working last call process. Finally, the authors would like to 2849 thank Ed Kern, the TEWG co-chair, for his comments and support. 2851 12.0 References 2853 [ASH1] J. Ash, M. Girish, E. Gray, B. Jamoussi, G. Wright, 2854 "Applicability Statement for CR-LDP," Work in Progress, July 2000. 2856 [ASH2] J. Ash, Dynamic Routing in Telecommunications Networks, McGraw 2857 Hill, 1998 2859 [ASH3] J. Ash, "TE & QoS Methods for IP-, ATM-, & TDM-Based 2860 Networks," Work in Progress, Mar. 2001. 2862 [AWD1] D. Awduche and Y. Rekhter, "Multiprocotol Lambda Switching: 2863 Combining MPLS Traffic Engineering Control with Optical 2864 Crossconnects", IEEE Communications Magazine, March 2001. 2866 [AWD2] D. Awduche, "MPLS and Traffic Engineering in IP Networks," 2867 IEEE Communications Magazine, Dec. 1999. 2869 [AWD3] D. Awduche, L. Berger, D. Gan, T. Li, G. Swallow, and V. 2870 Srinivasan, "RSVP-TE: Extensions to RSVP for LSP Tunnels," Work in 2871 Progress, Feb. 2001. 2873 [AWD4] D. Awduche, A. Hannan, X. Xiao, " Applicability Statement for 2874 Extensions to RSVP for LSP-Tunnels," Work in Progress, Apr. 2000. 2876 [AWD5] D. Awduche et al, "An Approach to Optimal Peering Between 2877 Autonomous Systems in the Internet," International Conference on 2878 Computer Communications and Networks (ICCCN'98), Oct. 1998. 2880 [CRUZ] R. L. Cruz, "A Calculus for Network Delay, Part II: Network 2881 Analysis," IEEE Transactions on Information Theory, vol. 37, pp. 2882 132-141, 1991. 2884 [DIFF-TE] F. Le Faucheur, et al, "Requirements for support of Diff- 2885 Serv-aware MPLS Traffic Engineering", Work in Progress, May 2001. 2887 [ELW95] A. Elwalid, D. Mitra and R.H. Wentworth, "A New Approach for 2888 Allocating Buffers and Bandwidth to Heterogeneous, Regulated Traffic 2889 in an ATM Node," IEEE IEEE Journal on Selected Areas in 2890 Communications, 13:6, pp. 1115-1127, Aug. 1995. 2892 [FGLR] A. Feldmann, A. Greenberg, C. Lund, N. Reingold, and J. 2893 Rexford, "NetScope: Traffic Engineering for IP Networks," IEEE 2894 Network Magazine, 2000. 2896 [FLJA93] S. Floyd and V. Jacobson, "Random Early Detection Gateways 2897 for Congestion Avoidance," IEEE/ACM Transactions on Networking, Vol. 2898 1 Nov. 4., p. 387-413, Aug. 1993. 2900 [FLOY94] S. Floyd, "TCP and Explicit Congestion Notification," ACM 2901 Computer Communication Review, V. 24, No. 5, p. 10-23, Oct. 1994. 2903 [HUSS87] B.R. Hurley, C.J.R. Seidl and W.F. Sewel, "A Survey of 2904 Dynamic Routing Methods for Circuit-Switched Traffic," IEEE 2905 Communication Magazine, Sep. 1987. 2907 [ITU-E600] ITU-T Recommendation E.600, "Terms and Definitions of 2908 Traffic Engineering," Mar. 1993. 2910 [ITU-E701] ITU-T Recommendation E.701, "Reference Connections for 2911 Traffic Engineering," Oct. 1993. 2913 [ITU-E801] ITU-T Recommendation E.801, "Framework for Service Quality 2914 Agreement," Oct. 1996. 2916 [JAM] B. Jamoussi, "Constraint-Based LSP Setup using LDP," Work in 2917 Progress, Feb. 2001. 2919 [KATZ] D. Katz, D. Yeung, and K. Kompella, "Traffic Engineering 2920 Extensions to OSPF," Work in Progress, Feb. 2001. 2922 [LNO96] T. Lakshman, A. Neidhardt, and T. Ott, "The Drop from Front 2923 Strategy in TCP over ATM and its Interworking with other Control 2924 Features," Proc. INFOCOM'96, p. 1242-1250, 1996. 2926 [MA] Q. Ma, "Quality of Service Routing in Integrated Services 2927 Networks," PhD Dissertation, CMU-CS-98-138, CMU, 1998. 2929 [MATE] A. Elwalid, C. Jin, S. Low, and I. Widjaja, "MATE: MPLS 2930 Adaptive Traffic Engineering," Proc. INFOCOM'01, Apr. 2001. 2932 [MCQ80] J.M. McQuillan, I. Richer, and E.C. Rosen, "The New Routing 2933 Algorithm for the ARPANET," IEEE. Trans. on Communications, vol. 28, 2934 no. 5, pp. 711-719, May 1980. 2936 [MPLS-DIFF] F. Le Faucheur, et al, "MPLS Support of Differentiated 2937 Services", Work in Progress, February 2001. 2939 [MR99] D. Mitra and K.G. Ramakrishnan, "A Case Study of Multiservice, 2940 Multipriority Traffic Engineering Design for Data Networks," Proc. 2941 Globecom'99, Dec 1999. 2943 [RFC-1349] P. Almquist, "Type of Service in the Internet Protocol 2944 Suite," RFC 1349, Jul. 1992. 2946 [RFC-1458] R. Braudes, S. Zabele, "Requirements for Multicast 2947 Protocols," RFC 1458, May 1993. 2949 [RFC-1771] Y. Rekhter and T. Li, "A Border Gateway Protocol 4 (BGP- 2950 4)," RFC 1771, Mar. 1995. 2952 [RFC-1812] F. Baker (Editor), "Requirements for IP Version 4 2953 Routers," RFC 1812, Jun. 1995. 2955 [RFC-1992] I. Castineyra, N. Chiappa, and M. Steenstrup, "The Nimrod 2956 Routing Architecture," RFC 1992, Aug. 1996. 2958 [RFC-1997] R. Chandra, P. Traina, and T. Li, "BGP Community 2959 Attributes" RFC 1997, Aug. 1996. 2961 [RFC-1998] E. Chen and T. Bates, "An Application of the BGP Community 2962 Attribute in Multi-home Routing," RFC 1998, Aug. 1996. 2964 [RFC-2178] J. Moy, "OSPF Version 2," RFC 2178, July 1997. 2966 [RFC-2205] R. Braden, et. al., "Resource Reservation Protocol (RSVP) 2967 - Version 1 Functional Specification," RFC 2205, Sep. 1997. 2969 [RFC-2211] J. Wroclawski, "Specification of the Controlled-Load 2970 Network Element Service," RFC 2211, Sep. 1997. 2972 [RFC-2212] S. Shenker, C. Partridge, R. Guerin, "Specification of 2973 Guaranteed Quality of Service," RFC 2212, Sep. 1997 2975 [RFC-2215] S. Shenker and J. Wroclawski, "General Characterization 2976 Parameters for Integrated Service Network Elements," RFC 2215, Sep. 2977 1997. 2979 [RFC-2216] S. Shenker and J. Wroclawski, "Network Element Service 2980 Specification Template," RFC 2216, Sep. 1997. 2982 [RFC-2330] V. Paxson et al., "Framework for IP Performance Metrics," 2983 RFC 2330, May 1998. 2985 [RFC-2386] E. Crawley, R. Nair, B. Rajagopalan, and H. Sandick, "A 2986 Framework for QoS-based Routing in the Internet," RFC 2386, Aug. 2987 1998. 2989 [RFC-2475] S. Blake et al., "An Architecture for Differentiated 2990 Services," RFC 2475, Dec. 1998. 2992 [RFC-2597] J. Heinanen, F. Baker, W. Weiss, and J. Wroclawski, 2993 "Assured Forwarding PHB Group," RFC 2597, June 1999. 2995 [RFC-2678] J. Mahdavi and V. Paxson, "IPPM Metrics for Measuring 2996 Connectivity," RFC 2678, Sep. 1999. 2998 [RFC-2679] G. Almes, S. Kalidindi, and M. Zekauskas, "A One-way Delay 2999 Metric for IPPM," RFC 2679, Sep. 1999. 3001 [RFC-2680] G. Almes, S. Kalidindi, and M. Zekauskas, "A One-way 3002 Packet Loss Metric for IPPM," RFC 2680, Sep. 1999. 3004 [RFC-2702] D. Awduche, J. Malcolm, J. Agogbua, M. O'Dell, J. McManus, 3005 "Requirements for Traffic Engineering over MPLS," RFC 2702, Sep. 3006 1999. 3008 [RFC-2722] N. Brownlee, C. Mills, and G. Ruth, "Traffic Flow 3009 Measurement: Architecture," RFC 2722, Oct. 1999. 3011 [RFC-2753] R. Yavatkar, D. Pendarakis, and R. Guerin, "A Framework 3012 for Policy-based Admission Control," RFC 2753, Jan. 2000. 3014 [RFC-2961] L. Berger, D. Gan, G. Swallow, P. Pan, F. Tommasi, S. 3015 Molendini, "RSVP Refresh Overhead Reduction Extensions", RFC 2961, 3016 Apr. 2000. 3018 [RFC-2998] Y. Bernet, et. al., "A Framework for Integrated Services 3019 Operation over Diffserv Networks", RFC 2998, Nov. 2000. 3021 [RFC-3031] E. Rosen, A. Viswanathan, R. Callon, "Multiprotocol Label 3022 Switching Architecture," RFC 3031, Jan. 2001. 3024 [RFC-3086] K. Nichols and B. Carpenter, "Definition of Differentiated 3025 Services Per Domain Behaviors and Rules for their Specification," RFC 3026 3086, April 2001. 3028 [RFC-3124] H. Balakrishnan and S. Seshan, "The Congestion Manager," 3029 RFC 3124, Jun. 2001. 3031 [SHAR] V. Sharma, et. al., "Framework for MPLS Based Recovery," Work 3032 in Progress, Mar. 2001. 3034 [SLDC98] B. Suter, T. Lakshman, D. Stiliadis, and A. Choudhury, 3035 "Design Considerations for Supporting TCP with Per-flow Queueing," 3036 Proc. INFOCOM'98, p. 299-306, 1998. 3038 [SMIT] H. Smit and T. Li, "IS-IS extensions for Traffic Engineering," 3039 Work in Progress, Feb. 2001. 3041 [XIAO] X. Xiao, A. Hannan, B. Bailey, L. Ni, "Traffic Engineering 3042 with MPLS in the Internet," IEEE Network magazine, Mar. 2000. 3044 [YARE95] C. Yang and A. Reddy, "A Taxonomy for Congestion Control 3045 Algorithms in Packet Switching Networks", IEEE Network Magazine, p. 3046 34-45, 1995. 3048 13.0 Authors' Addresses: 3050 Daniel O. Awduche 3051 Movaz Networks 3052 7926 Jones Branch Drive, Suite 615 3053 McLean, VA 22102 3054 Phone: 703-847-7350 3055 Email: awduche@movaz.com 3057 Angela Chiu 3058 Celion Networks 3059 1 Shiela Dr., Suite 2 3060 Tinton Falls, NJ 07724 3061 Phone: 732-747-9987 3062 Email: angela.chiu@celion.com 3064 Anwar Elwalid 3065 Lucent Technologies 3066 Murray Hill, NJ 07974 3067 Phone: 908 582-7589 3068 Email: anwar@lucent.com 3070 Indra Widjaja 3071 Bell Labs, Lucent Technologies 3072 600 Mountain Avenue 3073 Murray Hill, NJ 07974 3074 Phone: 908 582-0435 3075 Email: iwidjaja@research.bell-labs.com 3077 XiPeng Xiao 3078 Photuris Inc. 3079 2025 Stierlin Ct., 3080 Mountain View, CA 94043 3081 Phone: 650-919-3215 3082 Email: xxiao@photuris.com