idnits 2.17.1 draft-ietf-bmwg-ospfconv-intraarea-08.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 Internet-Drafts being working documents. ** The document seems to lack a 1id_guidelines paragraph about 6 months document validity. == Mismatching filename: the document gives the document name as 'draft-bmwg-ospfconv-intraarea-08', but the file name used is 'draft-ietf-bmwg-ospfconv-intraarea-08' ** The document is more than 15 pages and seems to lack a Table of Contents. == No 'Intended status' indicated for this document; assuming Proposed Standard == The page length should not exceed 58 lines per page, but there was 35 longer pages, the longest (page 2) being 60 lines == It seems as if not all pages are separated by form feeds - found 0 form feeds but 36 pages 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.) Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the RFC 3978 Section 5.4 Copyright Line does not match the current year == Couldn't figure out when the document was first submitted -- there may comments or warnings related to the use of a disclaimer for pre-RFC5378 work that could not be issued because of this. Please check the Legal Provisions document at https://trustee.ietf.org/license-info to determine if you need the pre-RFC5378 disclaimer. -- The document date (May 2004) is 7286 days in the past. Is this intentional? 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: 'OSPF' is mentioned on line 1411, but not defined == Missing Reference: 'TERM' is mentioned on line 731, but not defined == Missing Reference: 'RFC2328' is mentioned on line 236, but not defined == Outdated reference: A later version (-07) exists of draft-ietf-bmwg-ospfconv-applicability-05 ** Downref: Normative reference to an Informational draft: draft-ietf-bmwg-ospfconv-applicability (ref. 'APPLICABILITY') -- No information found for draft-bmwg-ospfconv-intraarea - is the name correct? -- Possible downref: Normative reference to a draft: ref. 'BENCHMARK' == Outdated reference: A later version (-06) exists of draft-ietf-isis-igp-p2p-over-lan-03 == Outdated reference: A later version (-09) exists of draft-ietf-ospf-scalability-06 Summary: 6 errors (**), 0 flaws (~~), 12 warnings (==), 3 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Network Working Group Vishwas Manral 3 Internet Draft Netplane Systems 4 Russ White 5 Cisco Systems 6 Aman Shaikh 7 Expiration Date: Novemeber 2004 University of California 8 File Name: draft-bmwg-ospfconv-intraarea-08.txt May 2004 10 Benchmarking Basic OSPF Single Router Control Plane Convergence 11 draft-ietf-bmwg-ospfconv-intraarea-08.txt 13 1. Status of this Memo 15 This document is an Internet-Draft and is in full conformance with 16 all provisions of Section 10 of RFC2026. 18 Internet Drafts are working documents of the Internet Engineering 19 Task Force (IETF), its Areas, and its Working Groups. Note that other 20 groups may also distribute working documents as Internet Drafts. 22 Internet Drafts are draft documents valid for a maximum of six 23 months. Internet Drafts may be updated, replaced, or obsoleted by 24 other documents at any time. It is not appropriate to use Internet 25 Drafts as reference material or to cite them other than as a "working 26 draft" or "work in progress". 28 The list of current Internet-Drafts can be accessed at 29 http://www.ietf.org/ietf/1id-abstracts.txt 31 The list of Internet-Draft Shadow Directories can be accessed at 32 http://www.ietf.org/shadow.html. 34 Copyright Notice 36 Copyright (C) The Internet Society (2002). All Rights Reserved. 38 Abstract 40 This draft provides suggestions for measuring OSPF single router 41 control plane convergence. Its initial emphasis is on the control 42 plane of single OSPF routers. We do not address forwarding plane 43 performance. 45 NOTE: Within this document, the word convergence relates to single 46 router control plane convergence only. 48 2. Introduction 50 There is a growing interest in routing protocol convergence testing, 51 with many people looking at various tests to determine how long it 52 takes for a network to converge after various conditions occur. The 53 major problem with this sort of testing is that the framework of the 54 tests has a major impact on the results; for instance, determining 55 when a network is converged, what parts of the router's operation are 56 considered within the testing, and other such things will have a 57 major impact on what apparent performance routing protocols provide. 59 This document attempts to provide a framework within which Open 60 Shortest Path First [OSPF] performance testing can be placed, and 61 provide some tests with which some aspects of OSPF performance can be 62 measured. The motivation of the draft is to provide a set of tests 63 that can provide the user comparable data from various vendors with 64 which to evaluate the OSPF protocol performance on the devices. 66 3. Specification of Requirements 68 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 69 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 70 document are to be interpreted as described in [RFC2119]. 72 4. Overview & Scope 74 While this document describes a specific set of tests aimed at 75 characterizing the single router control plane convergence 76 performance of OSPF processes in routers or other boxes that 77 incorporate OSPF functionality, a key objective is to propose 78 methodologies that will prdouce directly comparable convergence 79 related measurements. 81 Things which are outside the scope of this document include: 83 o The interactions of convergence and forwarding; testing is res- 84 tricted to events occurring within the control plane. Forwarding 85 performance is the primary focus in [INTERCONNECT] and it is 86 expected to be dealt with in work that ensues from [FIB-TERM]. 88 o Inter-area route generation, AS-external route generation, and 89 simultaneous traffic on the control and data paths within the 90 DUT. While the tests outlined in this document measure SPF time, 91 flooding times, and other aspects of all OSPF convergence per- 92 formance, it does not provide tests for measuring external or 93 summary route generation, route translation, or other OSPF 94 inter-area and external routing performance. These are expected 95 to be dealt with in a later draft. 97 Other drafts in the future may cover some of the items noted as not 98 covered in the scope of this draft. For a discussion of the terminol- 99 ogy used in this draft (in relation to the tests themselves), refer 100 to [TERM]. For a discussion of the applicability of this draft, refer 101 to [APPLICABILITY]. 103 While this draft assumes OSPFv2, which only carries routing informa- 104 tion for IPv4 destinations, the tests described in this document 105 apply to OSPFv3, which carries IPv6 destinations. 107 5. Test Conditions 109 In all tests, the following test conditions will be assumed: 111 o The link speed should be high enough so that does not become a 112 bottleneck. Link speeds of 10MBps or higher are recommended. The 113 link speed between routers should be specified in the test 114 report. 116 o For all point-to-point links, it is assumed that a link failure 117 results in an immediate notification to the operating system, 118 and thus to the OSPF process; this is explained thoroughly in 119 [MILLISEC]. 121 o No data traffic will be running between the routers during these 122 tests. 124 o Optional capabilities which can reduce performance, such as 125 authentication, should be noted in the test results if they are 126 enabled. 128 o Optional changes in the default timer values, such as the SPF, 129 hello, router dead, and other intervals, should be noted in the 130 test results. 132 o All places where injecting a set of LSAs is referenced, the set 133 can include varying numbers of LSAs of varying types represent- 134 ing a varying number of reachable destinations. See [TERM] for 135 further information about issues with LSA sets and network topo- 136 logies. 138 Tests should be run more than once, since a single test run 139 cannot be relied on to produce statistically sound results. The 140 number of test runs and any variations between the tests should 141 be recorded in the test results (see [TERM] for more information 142 on what items should be recorded in the test results). 144 6. Reference Topologies 146 Several reference topologies will be used throughout the tests 147 described in the remainder of this document. Rather than repeating 148 these topologies, we've gathered them all in one section. 150 o Reference Topology 1 (Emulated Topology) 152 ( ) 153 DUT----Generator----( emulated topology ) 154 ( ) 156 A simple back-to-back configuration. It's assumed that the link 157 between the generator and the DUT is a point-to-point link, 158 while the connections within the generator represent some emu- 159 lated topology. 161 o Reference Topology 2 (Generator and Collector) 163 ( ) 164 Collector-----DUT-----Generator--( emulated topology ) 165 \ / ( ) 166 \------------/ 168 All routers are connected through point-to-point links. The cost 169 of all links is assumed to be the same unless otherwise noted. 171 o Reference Topology 3 (Broadcast Network) 173 DUT R1 R2 174 | | | 175 -+------+------+-----..... 177 Any number of routers could be included on the common broadcast 178 network. 180 o Reference Topology 4 (Parallel Links) 182 /--(link 1)-----\ ( ) 183 DUT Generator--( emulated topology ) 184 \--(link 2)-----/ ( ) 186 In all cases the tests and topologies are designed to allow perfor- 187 mance measurements to be taken all on a single device, whether the 188 DUT or some other device in the network. This eliminates the need for 189 syncronized clocks within the test networks. 191 7. Basic Process Performance Tests 193 These tests will measure aspects of the OSPF implementation as a pro- 194 cess on the device under test, including: 196 o Time required to process an LSA 198 o Flooding time 200 o Shortest Path First computation 202 7.1. Time required to process an LSA 204 o Using reference topology 1 (Emulated Topology), begin with all 205 links up and a full adjacency established between the DUT and 206 the generator. 208 Note: The generator does not have direct knowledge of the state 209 of the adjacency on the DUT. The fact the adjacency may be in 210 Full on the generator does not mean that the DUT is ready. It 211 may still (and is likely to) be requesting LSAs from the genera- 212 tor. This process, involving processing of requested LSAs, will 213 affect the results of the test. The generator should either wait 214 until it sees the DUT's router-LSA listing the adjacency with 215 the generator or introduce a configurable delay before starting 216 the test. 218 o Send an LSA that is already there in the DUT (a duplicate LSA), 219 note the time difference between when the LSA is sent to when 220 the ack is received. This measures the time to propagate the LSA 221 and the ack, as well as processing time of the duplicate LSA. 222 This is dupLSAprocTime. 224 o Send a new LSA from the generator to the DUT, followed immedi- 225 ately by a duplicate LSA (LSA that already resides in the data- 226 base of DUT, but not the same as the one just sent). 228 o The DUT will acknowledge this second LSA immediately; note the 229 time of this acknowledgement. This is newLSAprocTime. 231 The amount of time required for an OSPF implementation to pro- 232 cess the new LSA can be computed by subtracting dupLSAprocTime 233 from newLSAprocTime. 235 Note: The duplicate LSA cannot be the same as the one just sent 236 because of the MinLSInterval restriction.[RFC2328] This test is 237 taken from [BLACKBOX]. 239 7.2. Flooding Time 241 o Using reference topology 2 (Generator and Collector), enable 242 OSPF on all links and allow the devices to build full adjacen- 243 cies. Configure the collector so it will block all flooding 244 towards the DUT, although it continues receiving advertisements 245 from the DUT. 247 o Inject a new set of LSAs from the generator towards the collec- 248 tor and the DUT. 250 o On the collector, note the time the flooding is complete across 251 the link to the generator. Also note the time the flooding is 252 complete across the link from the DUT. 254 The time between the last LSA is received on the collector from the 255 generator and the time the last LSA is received on the collector from 256 the DUT should be measured during this test. This time is important 257 in link state protocols, since the loop free nature of the network is 258 reliant on the speed at which revised topology information is 259 flooded. 261 Depending on the number of LSAs flooded, the sizes of the LSAs, the 262 number of LSUs, and the rate of flooding, these numbers could vary by 263 some amount. The settings and variances of these numbers should be 264 reported with the test results. 266 7.3. Shortest Path First Computation Time 268 o Use reference topology 1 (Emulated Toplogy), beginning with the 269 DUT and the generator fully adjacent. 271 o The default SPF timer on the DUT should be set to 0, so that any 272 new LSA that arrives, immediately results in the SPF calculation 274 [BLACKBOX]. 276 o The generator should inject a set of LSAs towards the DUT; the 277 DUT should be allowed to converge and install all best paths in 278 the local routing table, etc.. 280 o Send an LSA that is already there in the DUT (a duplicate LSA), 281 note the time difference between when the LSA is sent to when 282 the ack is received. This measures the time to propagate the LSA 283 and the ack, as well as processing time of the duplicate LSA. 284 This is dupLSAprocTime. 286 o Change the link cost between the generator and the emulated net- 287 work it is advertising, and transmit the new LSA to the DUT. 289 o Immediately inject another LSA which is a duplicate of some 290 other LSA the generator has previously injected (preferrably a 291 stub network someplace within the emulated network). 293 Note: The generator should make sure that outbound LSA packing 294 is not performed for the duplicate LSAs and they are always sent 295 in a separate Link-state Update packet. Otherwise, if the LSA 296 carrying the topo change and the duplicate LSA are in the same 297 packet, the SPF will be started the duplicate LSA is acked. 299 o Measure the time between transmitting the second (duplicate) LSA 300 and the acknowledgement for that LSA; this is the totalSPFtime. 301 The total time required to run SPF can be computed by subtract- 302 ing dupLSAprocTime from totalSPFtime. 304 The accuracy of this test is crucially dependant on the amount of 305 time between the transmission of the first and second LSAs. If there 306 is too much time between them, the test is meaningless because the 307 SPF run will complete before the second (duplicate) LSA is received. 308 If there is too little time between the LSAs being generated, then 309 they will both be handled before the SPF run is scheduled and 310 started, and thus the measurement would only be for the handling of 311 the duplicate LSA. 313 This test is also specified in [BLACKBOX]. 315 Note: This test may not be accurate on systems which implement OSPF 316 as a multithreaded process, where the flooding takes place in a 317 separate process (or on a different processor) than shortest path 318 first computations. 320 It is also possible to measure the SPF time using white box tests 321 (using output supplied by the OSPF software impelemtor). For 322 instance: 324 o Using reference topology 1 (Emulated Topology), establish a full 325 adjacency between the generator and the DUT. 327 o Inject a set of LSAs from the generator towards the DUT. Allow 328 the DUT to stabilize and install all best paths in the routing 329 table, etc. 331 o Change the link cost between the DUT and the generator (or the 332 link between the generator and the emulated network it is 333 advertising), such that a full SPF is required to run, although 334 only one piece of information is changed. 336 o Measure the amount of time required for the DUT to compute new 337 shortest path tree as a result of the topology changes injected 338 by the generator. These measurements should be taken using 339 available show and debug information on the DUT. 341 Several caveats MUST be mentioned when using a white box method of 342 measuring SPF time; for instance, such white box tests are only 343 applicable when testing various versions or variations within a sin- 344 gle implementation of the OSPF protocol. Futher, the same set of com- 345 mands MUST be used in each iteration of such a test, to ensure con- 346 sistent results. 348 There is some interesting relationship between the SPF times reported 349 by white box (internal) testing, and black box (external) testing; 350 these two types of tests may be used as a "sanity check" on the other 351 type of tests, by comparing the results of the two tests. 353 See [APPLICABILITY] for further discussion. 355 8. Basic Intra-Area OSPF tests 357 These tests measure the performance of an OSPF implementation for 358 basic intra-area tasks, including: 360 o Forming Adjacencies on Point-to-Point Link (Initialization) 362 o Forming Adjacencies on Point-to-Point Links 364 o Link Up with Information Already in the Database 365 o Initial convergence Time on a Designated Router Electing (Broad- 366 cast) Network 368 o Link Down with Layer 2 Detection 370 o Link Down with Layer 3 Detection 372 o Designated Router Election Time on A Broadcast Network 374 8.1. Forming Adjacencies on Point-to-Point Link (Initialization) 376 This test measures the time required to form an OSPF adjacency from 377 the time a layer two (data link) connection is formed between two 378 devices running OSPF. 380 o Use reference topology 1 (Emulated Topology), beginning with the 381 link between the generator and DUT disabled on the DUT. OSPF 382 should be configured and operating on both devices. 384 o Inject a set of LSAs from the generator towards the DUT. 386 o Bring the link up at the DUT, noting the time that the link car- 387 rier is established on the generator. 389 o Note the time the acknowledgement for the last LSA transmitted 390 from the DUT is received on the generator. 392 The time between the carrier establishment and the acknowledgement 393 for the last LSA transmitted by the generator should be taken as the 394 total amount of time required for the OSPF process on the DUT to 395 react to a link up event with the set of LSAs injected, including the 396 time required for the operating system to notify the OSPF process 397 about the link up, etc.. The acknowledgement for the last LSA 398 transmitted is used instead of the last acknowledgement received in 399 order to prevent timing skews due to retransmitted acknowledgements 400 or LSAs. 402 8.2. Forming Adjacencies on Point-to-Point Links 404 This test measures the time required to form an adjacency from the 405 time the first communication occurs between two devices running OSPF. 407 o Using reference topology 1 (Emulated Topology), configure the 408 DUT and the generator so traffic can be passed along the link 409 between them. 411 o Configure the generator so OSPF is running on the point-to-point 412 link towards the DUT, and inject a set of LSAs. 414 o Configure the DUT so OSPF is initialized, but not running on the 415 point-to-point link between the DUT and the generator. 417 o Enable OSPF on the interface between the DUT and the generator 418 on the DUT. 420 o Note the time of the first hello received from the DUT on the 421 generator. 423 o Note the time of the acknowledgement from the DUT for the last 424 LSA transmitted on the generator. 426 The time between the first hello received and the acknowledgement for 427 the last LSA transmitted by the generator should be taken as the 428 total amount of time required for the OSPF process on the DUT to 429 build a FULL neighbor adjacency with the set of LSAs injected. The 430 acknowledgement for the last LSA transmitted is used instead of the 431 last acknowledgement received in order to prevent timing skews due to 432 retransmitted acknowledgements or LSAs. 434 8.3. Forming adjacencies with Information Already in the Database 436 o Using reference topology 2 (Generator and Collector), configure 437 all three devices to run OSPF. 439 o Configure the DUT so the link between the DUT and the generator 440 is disabled . 442 o Inject a set of LSAs into the network from the generator; the 443 DUT should receive these LSAs through normal flooding from the 444 collector. 446 o Enable the link between the DUT and the generator. 448 o Note the time of the first hello received from the DUT on the 449 generator. 451 o Note the time of the last DBD received on the generator. 453 o Note the time of the acknowledgement from the DUT for the last 454 LSA transmitted on the generator. 456 The time between the hello received from the DUT by the generator and 457 the acknowledgement for the last LSA transmitted by the generator 458 should be taken as the total amount of time required for the OSPF 459 process on the DUT to build a FULL neighbor adjacency with the set of 460 LSAs injected. In this test, the DUT is already aware of the entire 461 network topology, so the time required should only include the pro- 462 cessing of DBDs exchanged when in EXCHANGE state, the time to build a 463 new router LSA containing the new connection information, and the 464 time required to flood and acknowledge this new router LSA. 466 The acknowledgement for the last LSA transmitted is used instead of 467 the last acknowledgement received in order to prevent timing skews 468 due to retransmitted acknowledgements or LSAs. 470 8.4. Designated Router Election Time on A Broadcast Network 472 o Using reference topology 3 (Broadcast Network), configure R1 to 473 be the designated router on the link, and the DUT to be the 474 backup designated router. 476 o Enable OSPF on the common broadcast link on all the routers in 477 the test bed. 479 o Disble the broadcast link on R1. 481 o Note the time of the last hello received from R1 on R2. 483 o Note the time of the first network LSA generated by the DUT as 484 received on R2. 486 The time between the last hello received on R2 and the first network 487 LSA generated by the DUT should be taken as the amount of time 488 required for the DUT to complete a designated router election compu- 489 tation. Note this test includes the dead interval timer at the DUT, 490 so this time may be factored out, or the hello and dead intervals 491 reduced to make these timers impact the overall test times less. All 492 changed timers, the number of routers connected to the link, and 493 other variable factors should be noted in the test results. 495 Note: If R1 sends a "goodbye hello," typically a hello with its 496 neighbor list empty, in the process of shutting down its interface, 497 using the time this hello is received instead of the time of the last 498 hello received would provide a more accurate measurement. 500 8.5. Initial Convergence Time on a Broadcast Network, Test 1 502 o Using reference topology 3 (Broadcast Network), begin with the 503 DUT connected to the network with OSPF enabled. OSPF should be 504 enabled on R1, but the broadcast link should be disabled. 506 o Enable the broadcast link between R1 and the DUT. Note the time 507 of the first hello received by R1. 509 o Note the time the first network LSA is flooded by the DUT at R1. 511 o The differential between the first hello and the first network 512 LSA is the time required by the DUT to converge on this new 513 topology. 515 This test assumes that the DUT will be the designated router on the 516 broadcast link. A similar test could be designed to test the conver- 517 gence time when the DUT is not the designated router as well. 519 This test may be performed with varying numbers of devices attached 520 to the broadcast network, and varying sets of LSAs being advertised 521 to the DUT from the routers attached to the broadcast network. Varia- 522 tions in the LSA sets and other factors should be noted in the test 523 results. 525 The time required to elect a designated router, as measured in Desig- 526 nated Router Election Time on A Broadcast Network, above, may be sub- 527 tracted from the results of this test to provide just the convergence 528 time across a broadcast network. 530 Note all the other tests in the document include route calculation 531 time in the conergence time, as described in [TERM], this test may 532 not include route calculation time in the resulting measured conver- 533 gence time, because initial route calculation may occur after the 534 first network LSA is flooded. 536 8.6. Initial Convergence Time on a Broadcast Network, Test 2 538 o Using reference topology 3 (Broadcast Network), begin with the 539 DUT connected to the network with OSPF enabled. OSPF should be 540 enabled on R1, but the broadcast link should be disabled. 542 o Enable the broadcast link between R1 and the DUT. Note the time 543 of the first hello transmitted by the DUT with a designated 544 router listed. 546 o Note the time the first network LSA is flooded by the DUT at R1. 548 o The differential between the first hello with a designated 549 router lists and the first network LSA is the time required by 550 the DUT to converge on this new topology. 552 8.7. Link Down with Layer 2 Detection 554 o Using reference topology 4 (Parallel Links), begin with OSPF in 555 the full state between the generator and the DUT. Both links 556 should be point-to-point links with the ability to notify the 557 operating system immediately upon link failure. 559 o Disable link 1; this should be done in such a way that the 560 keepalive timers at the data link layer will have no impact on 561 the DUT recognizing the link failure (the operating system in 562 the DUT should recognize this link failure immediately). Discon- 563 necting the cable on the generator end would be one possibility, 564 or shutting the link down. 566 o Note the time of the link failure on the generator. 568 o At the generator, note the time of the receipt of the new router 569 LSA from the DUT notifying the generator of the link 2 failure. 571 The difference in the time between the initial link failure and 572 the receipt of the LSA on the generator across link 2 should be 573 taken as the time required for an OSPF implementation to recog- 574 nize and process a link failure, including the time required to 575 generate and flood an LSA describing the link down event to an 576 adjacent neighbor. 578 8.8. Link Down with Layer 3 Detection 580 o Using reference topology 4 (Parallel Links), begin with OSPF in 581 the full state between the generator and the DUT. 583 o Disable OSPF processing on link 1 from the generator. This 584 should be done in such a way so it does not affect link status; 585 the DUT MUST note the failure of the adjacency through the dead 586 interval. 588 o At the generator, note the time of the receipt of the new router 589 LSA from the DUT notifying the generator of the link 2 failure. 591 The difference in the time between the initial link failure and the 592 receipt of the LSA on the generator across link 2 should be taken as 593 the time required for an OSPF implementation to recognize and process 594 an adjacency failure. 596 9. Security Considerations 598 This doecument does not modify the underlying security considerations 599 in [OSPF]. 601 10. Acknowledgements 603 Thanks to Howard Berkowitz, (hcb@clark.net), for his encouragement 604 and support. Thanks also to Alex Zinin (zinin@psg.net), Gurpreet 605 Singh (Gurpreet.Singh@SpirentCom.COM), and Yasuhiro Ohara 606 (yasu@sfc.wide.ad.jp) for their comments as well. 608 11. Normative References 610 [OPSF]Moy, J., "OSPF Version 2", RFC 2328, April 1998. 612 [TERM]Manral, V., "OSPF Convergence Testing Terminiology and Con- 613 cepts", draft-ietf-bmwg-ospfconv-term-08, May 2004 615 [APPLICABILITY] 616 Manral, V., "Benchmarking Applicability for Basic OSPF Conver- 617 gence", draft-ietf-bmwg-ospfconv-applicability-05, May 619 [RFC2119] 620 Bradner, S., "Key words for use in RFCs to Indicate Requirement 621 Levels", BCP 14, RFC 2119, March 1997 623 12. Informative References 625 [INTERCONNECT] 626 Bradner, S., McQuaid, J., "Benchmarking Methodology for Network 627 Interconnect Devices", RFC2544, March 1999. 629 [MILLISEC] 630 Alaettinoglu C., et al., "Towards Milli-Second IGP Convergence" 631 draft-alaettinoglu-isis-convergence 633 [FIB-TERM] 634 Trotter, G., "Terminology for Forwarding Information Base (FIB) 635 based Router Performance", RFC3222, October 2001. 637 [BLACKBOX] 638 Shaikh, Aman, Greenberg, Albert, "Experience in Black-Box OSPF 639 measurement" 641 13. Authors' Addresses 643 Vishwas Manral 644 Netplane Systems 645 189 Prashasan Nagar 646 Road number 72 647 Jubilee Hills 648 Hyderabad, India 650 vmanral@netplane.com 652 Russ White 653 Cisco Systems, Inc. 654 7025 Kit Creek Rd. 655 Research Triangle Park, NC 27709 657 riw@cisco.com 659 Aman Shaikh 660 AT&T Labs (Research) 661 180, Park Av 662 Florham Park, NJ 07932 664 ashaikh@research.att.com 666 Network Working Group Vishwas Manral 667 Internet Draft Netplane Systems 668 Russ White 669 Cisco Systems 670 Aman Shaikh 671 Expiration Date: November 2004 University of California 672 File Name: draft-ietf-bmwg-ospfconv-applicability-05.txt May 2004 674 Benchmarking Applicability for Basic OSPF Convergence 675 draft-ietf-bmwg-ospfconv-applicability-05.txt 677 Status of this Memo 679 This document is an Internet-Draft and is in full conformance with 680 all provisions of Section 10 of RFC2026. 682 Internet Drafts are working documents of the Internet Engineering 683 Task Force (IETF), its Areas, and its Working Groups. Note that other 684 groups may also distribute working documents as Internet Drafts. 686 Internet Drafts are draft documents valid for a maximum of six 687 months. Internet Drafts may be updated, replaced, or obsoleted by 688 other documents at any time. It is not appropriate to use Internet 689 Drafts as reference material or to cite them other than as a "working 690 draft" or "work in progress". 692 The list of current Internet-Drafts can be accessed at 693 http://www.ietf.org/ietf/1id-abstracts.txt 695 The list of Internet-Draft Shadow Directories can be accessed at 696 http://www.ietf.org/shadow.html. 698 Copyright Notice 700 Copyright (C) The Internet Society (2002). All Rights Reserved. 702 Abstract 704 This document discusses the applicability of various tests for 705 measuring single router control plane convergence, specifically in 706 regards to the Open Shortest First (OSPF) protocol. There are two 707 general sections in this document, the first discussing specific 708 advantages and limitations of specific OSPF convergence tests, and 709 the second discussing more general pitfalls to be considered when 710 testing routing protocols convergence testing. 712 1. Introduction 714 There is a growing interest in testing single router control plane 715 convergence for routing protocols, with many people looking at 716 testing methodologies which can provide information on how long it 717 takes for a network to converge after various network events occur. 718 It is important to consider the framework within which any given 719 convergence test is executed when attempting to apply the results of 720 the testing, since the framework can have a major impact on the 721 results. For instance, determining when a network is converged, what 722 parts of the router's operation are considered within the testing, 723 and other such things will have a major impact on what apparent 724 performance routing protocols provide. 726 This document describes in detail the various benefits and pitfalls 727 of tests described in [BENCHMARK]. It also explains how such 728 measurements can be useful for providers and the research community. 730 NOTE: The word convergence within this document refers to single 731 router control plane convergence [TERM]. 733 2. Advantages of Such Measurement 735 o To be able to compare the iterations of a protocol implementa- 736 tion. It is often useful to be able to compare the performance 737 of two iterations of a given implementation of a protocol to 738 determine where improvements have been made and where further 739 improvements can be made. 741 o To understand, given a set of parameters (network conditions), 742 how a particular implementation on a particular device is going 743 to perform. For instance, if you were trying to decide the pro- 744 cessing power (size of device) required in a certain location 745 within a network, you can emulate the conditions which are going 746 to exist at that point in the network and use the test described 747 to measure the perfomance of several different routers. The 748 results of these tests can provide one possible data point for 749 an intelligent decision. 751 If the device being tested is to be deployed in a running net- 752 work, using routes taken from the network where the equipment is 753 to be deployed rather than some generated topology in these 754 tests will give results which are closer to the real preformance 755 of the device. Care should be taken to emulate or take routes 756 from the actual location in the network where the device will be 757 (or would be) deployed. For instance, one set of routes may be 758 taken from an ABR, one set from an area 0 only router, various 759 sets from stub area, another set from various normal areas, etc. 761 o To measure the performance of an OSPF implementation in a wide 762 variety of scenarios. 764 o To be used as parameters in OSPF simulations by researchers. It 765 may some times be required for certain kinds of research to 766 measure the individual delays of each parameter within an OSPF 767 implementation. These delays can be measured using the methods 768 defined in [BENCHMARK]. 770 o To help optimize certain configurable parameters. It may some 771 times be helpful for operators to know the delay required for 772 individual tasks so as to optimize the resource usage in the 773 network i.e. if it is found that the processing time is x 774 seconds on an router, it would be helpful to determine the rate 775 at which to flood LSA's to that router so as to not overload the 776 network. 778 3. Assumptions Made and Limitations of such measurements 780 o The interactions of convergence and forwarding; testing is res- 781 tricted to events occurring within the control plane. Forwarding 782 performance is the primary focus in [INTERCONNECT] and it is 783 expected to be dealt with in work that ensues from [FIB-TERM]. 785 o Duplicate LSAs are Acknowledged Immediately. A few tests rely on 786 the property that duplicate LSA Acknowledgements are not delayed 787 but are done immediately. However if some implementation does 788 not acknowledge duplicate LSAs immediately on receipt, the test- 789 ing methods presented in [BENCHMARK] could give inaccurate meas- 790 urements. 792 o It is assumed that SPF is non-preemptive. If SPF is implemented 793 so that it can (and will be) preempted, the SPF measurements 794 taken in [BENCHMARK] would include the times that the SPF pro- 795 cess is not running ([BENCHMARK] measures the total time taken 796 for SPF to run, not the amount of time that SPF actually spends 797 on the device's processor), thus giving inaccurate measurements. 799 o Some implementations may be multithreaded or use a 800 multiprocess/multirouter model of OSPF. If because of this any 801 of the assumptions taken in measurement are violated in such a 802 model, it could lead to inaccurate measurements. 804 o The measurements resulting from the tests in [BENCHMARK] may not 805 provide the information required to deploy a device in a large 806 scale network. The tests described focus on individual com- 807 ponents of an OSPF implementation's performance, and it may be 808 difficult to combine the measurements in a way which accurately 809 depicts a device's performance in a large scale network. Further 810 research is required in this area. 812 o The measurements described in [BENCHMARK] should be used with 813 great care when comparing two different implementations of OSPF 814 from two different vendors. For instance, there are many other 815 factors than convergence speed that need to be taken into con- 816 sideration when comparing different vendor's products, and it's 817 difficult to align the resources available on one device to the 818 resources available on another device. 820 4. Observations on the Tests Described in [BENCHMARK] 822 Some observations taken while implementing the tests described in 823 [BENCHMARK] are noted in this section. 825 4.1. Measuring the SPF Processing Time Externally 827 The most difficult test to perform is the external measurement of the 828 time required to perform an SPF calculation, since the amount of time 829 between the first LSA which indicates a topology change and the 830 duplicate LSA is critical. If the duplicate LSA is sent too quickly, 831 it may be received before the device under test actually begins run- 832 ning SPF on the network change information. If the delay between the 833 two LSAs is too long, the device under test may finish SPF processing 834 before receiving the duplicate LSA. It is important to closely inves- 835 tigate any delays between the receipt of an LSA and the beginning of 836 an SPF calculation in the device under test; multiple tests with 837 various delays might be required to determine what delay needs to be 838 used to accurately measure the SPF calculation time. 840 Some implementations may force two intervals, the SPF hold time and 841 the SPF delay, between successive SPF calculations. If an SPF hold 842 time exists, it should be subtracted from the total SPF execution 843 time. If an SPF delay exists, it should be noted in the test results. 845 4.2. Noise in the Measurement Device 847 The device on which measurements are taken (not the device under 848 test) also adds noise to the test results, primarily in the form of 849 delay in packet processing and measurement output. The largest source 850 of noise is generally the delay between the receipt of packets by the 851 measuring device and the information about the packet reaching the 852 device's output, where the event can be measured. The following steps 853 may be taken to reduce this sampling noise: 855 o Increasing the number of samples taken will generally improve 856 the tester's ability to determine what is noise, and remove it 857 from the results. 859 o Try to take time-stamp for a packet as early as possible. 860 Depending on the operating system being used on the box, one can 861 instrument the kernel to take the time-stamp when the interrupt 862 is processed. This does not eliminate the noise completely, but 863 at least reduces it. 865 o Keep the measurement box as lightly loaded as possible. 867 o Having an estimate of noise can also be useful. 869 The DUT also adds noise to the measurement. Points (a) and (c) 870 apply to the DUT as well. 872 4.3. Gaining an Understanding of the Implementation Improves Measure- 873 ments 875 While the tester will (generally) not have access to internal infor- 876 mation about the OSPF implementation being tested using [BENCHMARK], 877 the more thorough the tester's knowledge of the implementation is, 878 the more accurate the results of the tests will be. For instance, in 879 some implementations, the installation of routes in local routing 880 tables may occur while the SPF is being calculated, dramatically 881 impacting the time required to calculate the SPF. 883 4.4. Gaining an Understanding of the Tests Improves Measurements 885 One method which can be used to become familiar with the tests 886 described in [BENCHMARK] is to perform the tests on an OSPF implemen- 887 tation for which all the internal details are available, such as 888 [GATED]. While there is no assurance that any two implementations 889 will be similar, this will provide a better understanding of the 890 tests themselves. 892 5. LSA and Destination mix 894 In many OSPF benchmark tests, a generator injecting a number of LSAs 895 is called for. There are several areas in which injected LSAs can be 896 varied in testing: 898 o The number of destinations represented by the injected LSAs 900 Each destination represents a single reachable IP network; these 901 will be leaf nodes on the shortest path tree. The primary impact 902 to performance should be the time required to insert destina- 903 tions in the local routing table and handling the memory 904 required to store the data. 906 o The types of LSAs injected 908 There are several types of LSAs which would be acceptable under 909 different situations; within an area, for instance, type 1, 2, 910 3, 4, and 5 are likely to be received by a router. Within a 911 not-so-stubby area, however, type 7 LSAs would replace the type 912 5 LSAs received. These sorts of characterizations are important 913 to note in any test results. 915 o The Number of LSAs injected 917 Within any injected set of information, the number of each type 918 of LSA injected is also important. This will impact the shortest 919 path algorithms ability to handle large numbers of nodes, large 920 shortest path first trees, etc. 922 o The Order of LSA Injection 924 The order in which LSAs are injected should not favor any given 925 data structure used for storing the LSA database on the device 926 under test. For instance, AS-External LSA's have AS wide flood- 927 ing scope; any Type-5 LSA originated is immediately flooded to 928 all neighbors. However the Type-4 LSA which announces the ASBR 929 as a border router is originated in an area at SPF time (by ABRs 930 on the edge of the area in which the ASBR is). If SPF isn't 931 scheduled immediately on the ABRs originating the type 4 LSA, 932 the type-4 LSA is sent after the type-5 LSA's reach a router in 933 the adjacent area. So routes to the external destinations aren't 934 immediately added to the routers in the other areas. When the 935 routers which already have the type 5's receive the type-4 LSA, 936 all the external routes are added to the tree at the same time. 937 This timing could produce different results than a router 938 receiving a type 4 indicating the presence of a border router, 939 followed by the type 5's originated by that border router. 941 The ordering can be changed in various tests to provide insight 942 on the efficiency of storage within the DUT. Any such changes in 943 ordering should be noted in test results. 945 6. Tree Shape and the SPF Algorithm 947 The complexity of Dijkstra's algorithm depends on the data structure 948 used for storing vertices with their current minimum distances from 949 the source, with the simplest structure being a list of vertices 950 currently reachable from the source. In a simple list of vertices, 951 finding the minimum cost vertex then would take O(size of the list). 952 There will be O(n) such operations if we assume that all the vertices 953 are ultimately reachable from the source. Moreover, after the vertex 954 with min cost is found, the algorithm iterates thru all the edges of 955 the vertex and updates cost of other vertices. With an adjacency list 956 representation, this step when iterated over all the vertices, would 957 take O(E) time, with E being the number of edges in the graph. Thus, 958 overall running time is: 960 O(sum(i:1, n)(size(list at level i) + E). 962 So, everything boils down to the size (list at level i). 964 If the graph is linear: 966 root 967 | 968 1 969 | 970 2 971 | 972 3 973 | 974 4 975 | 976 5 977 | 978 6 980 and source is a vertex on the end, then size(list at level i) = 1 981 for all i. Moreover, E = n - 1. Therefore, running time is O(n). 983 If graph is a balanced binary tree: 985 root 986 / \ 987 1 2 988 / \ / \ 989 3 4 5 6 991 size(list at level i) is a little complicated. First it increases 992 by 1 at each level upto a certain number, and then goes down by 1. 993 If we assumed that tree is a complete tree (like the one in the 994 draft) with k levels (1 to k), then size(list) goes on like this: 995 1, 2, 3, 997 Then the number of edges E is still n - 1. It then turns out that 998 the run-time is O(n^2) for such a tree. 1000 If graph is a complete graph (fully-connected mesh), then 1001 size(list at level i) = n - i. Number of edges E = O(n^2). There- 1002 fore, run-time is O(n^2). 1004 So, the performance of the shortest path first algorithm used to 1005 compute the best paths through the network is dependant o the con- 1006 struction of the tree The best practice would be to try and make 1007 any emulated network look as much like a real network as possible, 1008 especially in the area of the tree depth, the meshiness of the 1009 network, the number of stub links versus transit links, and the 1010 number of connections and nodes to process at each level within 1011 the original tree. 1013 7. Topology Generation 1015 As the size of networks grows, it becomes more and more difficult to 1016 actually create a large scale network on which to test the properties 1017 of routing protocols and their implementations. In general, network 1018 emulators are used to provide emulated topologies which can be adver- 1019 tised to a device with varying conditions. Route generators either 1020 tend to be a specialized device, a piece of software which runs on a 1021 router, or a process that runs on another operating system, such as 1022 Linux or another variant of Unix. 1024 Some of the characteristics of this device should be: 1026 o The ability to connect to the several devices using both point- 1027 to-point and broadcast high speed media. Point-to-point links 1028 can be emulated with high speed Ethernet as long as there is no 1029 hub or other device in between the DUT and the route generator, 1030 and the link is configured as a point-to-point link within OSPF 1031 [BROADCAST-P2P]. 1033 o The ability to create a set of LSAs which appear to be a logi- 1034 cal, realistic topology. For instance, the generator should be 1035 able to mix the number of point-to-point and broadcast links 1036 within the emulated topology, and should be able to inject vary- 1037 ing numbers of externally reachable destinations. 1039 o The ability to withdraw and add routing information into and 1040 from the emulated topology to emulate links flapping. 1042 o The ability to randomly order the LSAs representing the emulated 1043 topology as they are advertised. 1045 o The ability to log or otherwise measure the time between packets 1046 transmitted and received. 1048 o The ability to change the rate at which OSPF LSAs are transmit- 1049 ted. 1051 o The generator and the collector should be fast enough so that 1052 they are not bottle necks. The devices should also have a degree 1053 of granularity of measurement atleast as small as desired from 1054 the test results. 1056 8. Security Considerations 1058 This doecument does not modify the underlying security considerations 1059 in [OSPF]. 1061 9. Acknowledgements 1063 Thanks to Howard Berkowitz, (hcb@clark.net) and the rest of the BGP 1064 benchmarking team for their support and to Kevin 1065 Dubray(kdubray@juniper.net) who realized the need of this draft. 1067 10. Normative References 1069 [BENCHMARK] 1070 Manral, V., "Benchmarking Basic OSPF Single Router Control Plane 1071 Convergence", draft-bmwg-ospfconv-intraarea-08, May 2004 1073 [TERM]Manral, V., "OSPF Convergence Testing Terminiology and Con- 1074 cepts", draft-bmwg-ospfconv-term-08, May 2004 1076 [RFC2119] 1077 Bradner, S., "Key words for use in RFCs to Indicate Requirement 1078 Levels", BCP 14, RFC 2119, March 1997 1080 11. Informative References 1082 [INTERCONNECT] 1083 Bradner, S., McQuaid, J., "Benchmarking Methodology for Network 1084 Interconnect Devices", RFC2544, March 1999. 1086 [FIB-TERM] 1087 Trotter, G., "Terminology for Forwarding Information Base (FIB) 1088 based Router Performance", RFC3222, October 2001. 1090 [BROADCAST-P2P] 1091 Shen, Naiming, et al., "Point-to-point operation over LAN in 1092 link-state routing protocols", draft-ietf-isis-igp-p2p-over- 1093 lan-03.txt, August, 2003. 1095 [GATED] 1096 http://www.gated.org 1098 12. Authors' Addresses 1099 Vishwas Manral 1100 Netplane Systems 1101 189 Prashasan Nagar 1102 Road number 72 1103 Jubilee Hills 1104 Hyderabad, India 1106 vmanral@netplane.com 1108 Russ White 1109 Cisco Systems, Inc. 1110 7025 Kit Creek Rd. 1111 Research Triangle Park, NC 27709 1113 riw@cisco.com 1115 Aman Shaikh 1116 University of California 1117 School of Engineering 1118 1156 High Street 1119 Santa Cruz, CA 95064 1121 aman@soe.ucsc.edu 1123 Network Working Group Vishwas Manral 1124 Internet Draft Netplane Systems 1125 Russ White 1126 Cisco Systems 1127 Aman Shaikh 1128 Expiration Date: November 2004 University of California 1129 File Name: draft-bmwg-ospfconv-term-08.txt May 2004 1131 OSPF Benchmarking Terminology and Concepts 1132 draft-bmwg-ospfconv-term-06.txt 1134 Status of this Memo 1136 This document is an Internet-Draft and is in full conformance with 1137 all provisions of Section 10 of RFC2026. 1139 Internet Drafts are working documents of the Internet Engineering 1140 Task Force (IETF), its Areas, and its Working Groups. Note that other 1141 groups may also distribute working documents as Internet Drafts. 1143 Internet Drafts are draft documents valid for a maximum of six 1144 months. Internet Drafts may be updated, replaced, or obsoleted by 1145 other documents at any time. It is not appropriate to use Internet 1146 Drafts as reference material or to cite them other than as a "working 1147 draft" or "work in progress". 1149 The list of current Internet-Drafts can be accessed at 1150 http//www.ietf.org/ietf/1id-abstracts.txt 1152 The list of Internet-Draft Shadow Directories can be accessed at 1153 http//www.ietf.org/shadow.html. 1155 Copyright Notice 1157 Copyright (C) The Internet Society (2002). All Rights Reserved. 1159 Abstract 1161 This draft explains the terminology and concepts used in OSPF 1162 benchmarking. While some of these terms may be defined elsewhere, and 1163 we will refer the reader to those definitions in some cases, we also 1164 include discussions concerning these terms as they relate 1165 specifically to the tasks involved in benchmarking the OSPF protocol. 1167 1. Specification of Requirements 1169 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 1170 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 1171 document are to be interpreted as described in RFC 2119 [RFC2119]. 1173 2. Motivation 1175 This draft is a companion to [BENCHMARK], which describes basic Open 1176 Shortest Path First [OSPF] testing methods. This draft explains 1177 terminology and concepts used in OSPF Testing Framework Drafts, such 1178 as [BENCHMARK]. 1180 3. Common Definitions 1182 Definitions in this section are well known industry and benchmarking 1183 terms which may be defined elsewhere. 1185 o White Box (Internal) Measurements 1187 - Definition 1189 White Box measurements are measurements reported and col- 1190 lected on the Device Under Test (DUT) itself. 1192 - Discussion 1194 These measurement rely on output and event recording, along 1195 with the clocking and timestamping available on the DUT 1196 itself. Taking measurements on the DUT may impact the 1197 actual outcome of the test, since it can increase processor 1198 loading, memory utilization, and timing factors. Some dev- 1199 ices may not have the required output readily available for 1200 taking internal measurements, as well. 1202 Note: White box measurements can be influenced by the 1203 vendor's implementation of the various timers and process- 1204 ing models. Whenever possible, internal measurements should 1205 be compared to external measurements to verify and validate 1206 them. 1208 Because of the potential for variations in collection and 1209 presentation methods across different DUTs, white box 1210 measurements MUST NOT be used as a basis of comparison in 1211 benchmarks. This has been a guiding principal of Bench- 1212 marking Methodology Working Group. 1214 o Black Box (External) Measurements 1216 - Definition 1218 Black Box measurements infer the performance of the DUT 1219 through observation of its communications with other dev- 1220 ices. 1222 - Discussion 1224 One example of a black box measurement is when a downstream 1225 device receives complete routing information from the DUT, 1226 it can be inferred that the DUT has transmitted all the 1227 routing information available. External measurements of 1228 internal operations may suffer in that they include not 1229 just the protocol action times, but also propagation 1230 delays, queuing delays, and other such factors. 1232 For the purposes of [BENCHMARK], external techniques are 1233 more readily applicable. 1235 o Multi-device Measurements 1237 - Measurements assessing communications (usually in combina- 1238 tion with internal operations) between two or more DUTs. 1239 Multi-device measurements may be internal or external. 1241 4. Terms Defined Elsewhere 1243 Terms in this section are defined elsewhere, and included only to 1244 include a discussion of those terms in reference to [BENCHMARK]. 1246 o Point-to-Point links 1248 - Definition 1250 See [OSPF], Section 1.2. 1252 - Discussion 1254 A point-to-point link can take lesser time to converge than 1255 a broadcast link of the same speed because it does not have 1256 the overhead of DR election. Point-to-point links can be 1257 either numbered or unnumbered. However in the context of 1258 [BENCHMARK] and [OSPF], the two can be regarded the same. 1260 o Broadcast Link 1262 - Definition 1264 See [OSPF], Section 1.2. 1266 - Discussion 1268 The adjacency formation time on a broadcast link can be 1269 more than that on a point-to-point link of the same speed, 1270 because DR election has to take place. All routers on a 1271 broadcast network form adjacency with the DR and BDR. 1273 Async flooding also takes place thru the DR. In context of 1274 convergence, it may take more time for an LSA to be flooded 1275 from one DR-other router to another DR-other router, 1276 because the LSA has to be first processed at the DR. 1278 o Shortest Path First Execution Time 1280 - Definition 1281 The time taken by a router to complete the SPF process, as 1282 described in [OSPF]. 1284 - Discussion 1286 This does not include the time taken by the router to give 1287 routes to the forwarding engine. 1289 Some implementations may force two intervals, the SPF hold 1290 time and the SPF delay, between successive SPF calcula- 1291 tions. If an SPF hold time exists, it should be subtracted 1292 from the total SPF execution time. If an SPF delay exists, 1293 it should be noted in the test results. 1295 - Measurement Units 1297 The SPF time is generally measured in milliseconds. 1299 o Hello Interval 1301 - Definition 1303 See [OSPF], Section 7.1. 1305 - Discussion 1307 The hello interval should be the same for all routers on a 1308 network. 1310 Decreasing the hello interval can allow the router dead 1311 interval (below) to be reduced, thus reducing convergence 1312 times in those situations where the router dead interval 1313 timing out causes an OSPF process to notice an adjacency 1314 failure. Further discussion on small hello intervals is 1315 given in [OSPF-SCALING]. 1317 o Router Dead interval 1319 - Definition 1321 See [OSPF], Section 7.1. 1323 - Discussion 1325 This is advertised in the router's Hello Packets in the Router- 1326 DeadInterval field. The router dead interval should be some mul- 1327 tiple of the HelloInterval (say 4 times the hello interval), and 1328 must be the same for all routers attached to a common network. 1330 5. Concepts 1332 5.1. The Meaning of Single Router Control Plane Convergence 1334 A network is termed to be converged when all of the devices within 1335 the network have a loop free path to each possible destination. Since 1336 we are not testing network convergence, but performance for a partic- 1337 ular device within a network, however, this definition needs to be 1338 narrowed somewhat to fit within a single device view. 1340 In this case, convergence will mean the point in time when the DUT 1341 has performed all actions needed to react to the change in topology 1342 represented by the test condition; for instance, an OSPF device must 1343 flood any new information it has received, rebuild its shortest path 1344 first (SPF) tree, and install any new paths or destinations in the 1345 local routing information base (RIB, or routing table). 1347 Note that the word convergence has two distinct meanings; the process 1348 of a group of individuals meeting the same place, and the process of 1349 a single individual meeting in the same place as an existing group. 1350 This work focuses on the second meaning of the word, so we consider 1351 the time required for a single device to adapt to a network change to 1352 be Single Router Convergence. 1354 This concept does not include the time required for the control plane 1355 of the device to transfer the information required to forward packets 1356 to the data plane, nor the amount of time between the data plane 1357 receiving that information and being able to actually forward 1358 traffic. 1360 5.2. Measuring Convergence 1362 Obviously, there are several elements to convergence, even under the 1363 definition given above for a single device, including (but not lim- 1364 ited to): 1366 o The time it takes for the DUT to pass the information about a 1367 network event on to its neighbors. 1369 o The time it takes for the DUT to process information about a 1370 network event and calculate a new Shortest Path Tree (SPT). 1372 o The time it takes for the DUT to make changes in its local rib 1373 reflecting the new shortest path tree. 1375 5.3. Types of Network Events 1377 A network event is an event which causes a change in the network 1378 topology. 1380 o Link or Neighbor Device Up 1382 The time needed for an OSPF implementation to recoginize a new 1383 link coming up on the device, build any necessarily adjacencies, 1384 synchronize its database, and perform all other needed actions 1385 to converge. 1387 o Initialization 1389 The time needed for an OSPF implementation to be initialized, 1390 recognize any links across which OSPF must run, build any needed 1391 adjacencies, synchronize its database, and perform other actions 1392 needed to converge. 1394 o Adjacency Down 1396 The time needed for an OSPF implementation to recognize a link 1397 down/adjacency loss based on hello timers alone, propogate any 1398 information as necessary to its remaining adjacencies, and per- 1399 form other actions needed to converge. 1401 o Link Down 1403 The time needed for an OSPF implementation to recognize a link 1404 down based on layer 2 provided information, propogate any infor- 1405 mation as needed to its remaining adjacencies, and perform other 1406 actions needed to converge. 1408 6. Security Considerations 1410 This doecument does not modify the underlying security considerations 1411 in [OSPF]. 1413 7. Acknowedgements 1415 The authors would like to thank Howard Berkowitz (hcb@clark.net), 1416 Kevin Dubray, (kdubray@juniper.net), Scott Poretsky 1417 (sporetsky@avici.com), and Randy Bush (randy@psg.com) for their dis- 1418 cussion, ideas, and support. 1420 8. Normative References 1422 [BENCHMARK] 1423 Manral, V., "Benchmarking Basic OSPF Single Router Control Plane 1424 Convergence", draft-bmwg-ospfconv-intraarea-08, May 2004. 1426 [OSPF]Moy, J., "OSPF Version 2", RFC 2328, April 1998. 1428 9. Informative References 1430 [OSPF-SCALING] 1431 Choudhury, Gagan L., Editor, "Prioritized Treatment of Specific 1432 OSPF Packets and Congestion Avoidance", draft-ietf-ospf- 1433 scalability-06.txt, August 2003. 1435 10. Authors' Addresses 1437 Vishwas Manral, 1438 Netplane Systems, 1439 189 Prashasan Nagar, 1440 Road number 72, 1441 Jubilee Hills, 1442 Hyderabad. 1444 vmanral@netplane.com 1446 Russ White 1447 Cisco Systems, Inc. 1448 7025 Kit Creek Rd. 1450 Research Triangle Park, NC 27709 1452 riw@cisco.com 1454 Aman Shaikh 1455 University of California 1456 School of Engineering 1457 1156 High Street 1458 Santa Cruz, CA 95064 1460 aman@soe.ucsc.edu