idnits 2.17.1 draft-ietf-ippm-tcp-throughput-tm-13.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- No issues found here. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (May 31, 2011) is 4714 days in the past. Is this intentional? Checking references for intended status: Informational ---------------------------------------------------------------------------- ** Obsolete normative reference: RFC 1323 (Obsoleted by RFC 7323) Summary: 1 error (**), 0 flaws (~~), 1 warning (==), 1 comment (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 1 Network Working Group B. Constantine 2 Internet-Draft JDSU 3 Intended status: Informational G. Forget 4 Expires: November 30, 2011 Bell Canada (Ext. Consultant) 5 Ruediger Geib 6 Deutsche Telekom 7 Reinhard Schrage 8 Schrage Consulting 10 May 31, 2011 12 Framework for TCP Throughput Testing 13 draft-ietf-ippm-tcp-throughput-tm-13.txt 15 Abstract 17 This framework describes a practical methodology for measuring end- 18 to-end TCP Throughput in a managed IP network. The goal is to provide 19 a better indication in regards to user experience. In this framework, 20 TCP and IP parameters are specified to optimize TCP throughput. 22 Requirements Language 24 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 25 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 26 document are to be interpreted as described in RFC 2119 [RFC2119]. 28 Status of this Memo 30 This Internet-Draft is submitted in full conformance with the 31 provisions of BCP 78 and BCP 79. 33 Internet-Drafts are working documents of the Internet Engineering 34 Task Force (IETF). Note that other groups may also distribute 35 working documents as Internet-Drafts. The list of current Internet- 36 Drafts is at http://datatracker.ietf.org/drafts/current/. 38 Internet-Drafts are draft documents valid for a maximum of six months 39 and may be updated, replaced, or obsoleted by other documents at any 40 time. It is inappropriate to use Internet-Drafts as reference 41 material or to cite them other than as "work in progress." 43 This Internet-Draft will expire on November 30, 2011. 45 Copyright Notice 47 Copyright (c) 2011 IETF Trust and the persons identified as the 48 document authors. All rights reserved. 50 This document is subject to BCP 78 and the IETF Trust's Legal 51 Provisions Relating to IETF Documents 52 (http://trustee.ietf.org/license-info) in effect on the date of 53 publication of this document. Please review these documents 54 carefully, as they describe your rights and restrictions with respect 55 to this document. Code Components extracted from this document must 56 include Simplified BSD License text as described in Section 4.e of 57 the Trust Legal Provisions and are provided without warranty as 58 described in the Simplified BSD License. 60 Table of Contents 62 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 3 63 1.1 Terminology. . . . . . . . . . . . . . . . . . . . . . . . 4 64 1.2 TCP Equilibrium . . . . . . . . . . . . . . . . . . . . . 5 65 2. Scope and Goals . . . . . . . . . . . . . . . . . . . . . . . 6 66 3. Methodology. . . . . . . . . . . . . . . . . . . . . . . . . . 7 67 3.1 Path MTU . . . . . . . . . . . . . . . . . . . . . . . . . 9 68 3.2 Round Trip Time (RTT) and Bottleneck Bandwidth (BB). . . . 9 69 3.2.1 Measuring RTT . . . . . . . . . . . . . . . . . . . . 9 70 3.2.2 Measuring BB . . . . . . . . . . . . . . . . . . . . 10 71 3.3. Measuring TCP Throughput . . . . . . . . . . . . . . . . . 11 72 3.3.1 Minimum TCP RWND . . . . . . . . . . . . . . . . . . . 11 73 4. TCP Metrics . . . . . . . . . . . . . . . . . . . . . . . . . 14 74 4.1 Transfer Time Ratio. . . . . . . . . . . . . . . . . . . . 14 75 4.1.1 Maximum Achievable TCP Throughput calculation . . . . 15 76 4.1.2 Transfer Time and Transfer Time Ratio calculation. . . 16 77 4.2 TCP Efficiency . . . . . . . . . . . . . . . . . . . . . . 17 78 4.2.1 TCP Efficiency Percentage calculation . . . . . . . . 17 79 4.3 Buffer Delay . . . . . . . . . . . . . . . . . . . . . . . 17 80 4.3.1 Buffer Delay Percentage calculation. . . . . . . . . . 17 81 5. Conducting TCP Throughput Tests. . . . . . . . . . . . . . . . 18 82 5.1 Single versus Multiple Connections . . . . . . . . . . . . 18 83 5.2 Results Interpretation . . . . . . . . . . . . . . . . . . 19 84 6. Security Considerations . . . . . . . . . . . . . . . . . . . 21 85 7. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 21 86 8. Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . 22 87 9. References . . . . . . . . . . . . . . . . . . . . . . . . . . 22 88 9.1 Normative References . . . . . . . . . . . . . . . . . . . 22 89 9.2 Informative References . . . . . . . . . . . . . . . . . . 22 91 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 23 93 1. Introduction 95 In the network industry, the SLA (Service Level Agreement) provided 96 to business class customers is generally based upon Layer 2/3 97 criteria such as: Bandwidth, latency, packet loss and delay 98 variations (jitter). Network providers are coming to the realization 99 that Layer 2/3 testing is not enough to adequately ensure end-user's 100 satisfaction. In addition to Layer 2/3 testing, this framework 101 recommends a methodology for measuring TCP Throughput in order to 102 provide meaningful results with respect to user experience. 104 Additionally, business class customers seek to conduct repeatable TCP 105 Throughput tests between locations. Since these organizations rely on 106 the networks of the providers, a common test methodology with 107 predefined metrics would benefit both parties. 109 Note that the primary focus of this methodology is managed business 110 class IP networks; e.g. those Ethernet terminated services for which 111 organizations are provided an SLA from the network provider. Because 112 of the SLA, the expectation is that the TCP Throughput should achieve 113 the guaranteed bandwidth. End-users with "best effort" access could 114 use this methodology, but this framework and its metrics are intended 115 to be used in a predictable managed IP network. No end-to-end 116 performance can be guaranteed when only the access portion is being 117 provisioned to a specific bandwidth capacity. 119 The intent behind this document is to define a methodology for 120 testing sustained TCP Layer performance. In this document, the 121 achievable TCP Throughput is that amount of data per unit time that 122 TCP transports when in the TCP Equilibrium state. (See Section 1.2 123 for TCP Equilibrium definition). Throughout this document, maximum 124 achievable throughput refers to the theoretical achievable throughput 125 when TCP is in the Equilibrium state. 127 TCP is connection oriented and at the transmitting side it uses a 128 congestion window, (TCP CWND). At the receiving end, TCP uses a 129 receive window, (TCP RWND) to inform the transmitting end on how 130 many Bytes it is capable to accept at a given time. 132 Derived from Round Trip Time (RTT) and network Bottleneck Bandwidth 133 (BB), the Bandwidth Delay Product (BDP) determines the Send and 134 Received Socket buffers sizes required to achieve the maximum TCP 135 Throughput. Then, with the help of slow start and congestion 136 avoidance algorithms, a TCP CWND is calculated based on the IP 137 network path loss rate. Finally, the minimum value between the 138 calculated TCP CWND and the TCP RWND advertised by the opposite end 139 will determine how many Bytes can actually be sent by the 140 transmitting side at a given time. 142 Both TCP Window sizes (RWND and CWND) may vary during any given TCP 143 session, although up to bandwidth limits, larger RWND and larger CWND 144 will achieve higher throughputs by permitting more in-flight Bytes. 146 At both ends of the TCP connection and for each socket, there are 147 default buffer sizes. There are also kernel enforced maximum buffer 148 sizes. These buffer sizes can be adjusted at both ends (transmitting 149 and receiving). Some TCP/IP stack implementations use Receive Window 150 Auto-Tuning, although in order to obtain the maximum throughput it is 151 critical to use large enough TCP Send and Receive Socket Buffer 152 sizes. In fact, they SHOULD be equal to or greater than BDP. 154 Many variables are involved in TCP Throughput performance, but this 155 methodology focuses on: 156 - BB (Bottleneck Bandwidth) 157 - RTT (Round Trip Time) 158 - Send and Receive Socket Buffers 159 - Minimum TCP RWND 160 - Path MTU (Maximum Transmission Unit) 162 This methodology proposes TCP testing that SHOULD be performed in 163 addition to traditional Layer 2/3 type tests. In fact, Layer 2/3 164 tests are REQUIRED to verify the integrity of the network before 165 conducting TCP tests. Examples include iperf (UDP mode) and manual 166 packet layer test techniques where packet throughput, loss, and delay 167 measurements are conducted. When available, standardized testing 168 similar to [RFC2544] but adapted for use in operational networks MAY 169 be used. 171 Note: [RFC2544] was never meant to be used outside a lab environment. 173 Sections 2 and 3 of this document provides a general overview of the 174 proposed methodology. Section 4 defines the metrics while Section 5 175 explains how to conduct the tests and interpret the results. 177 1.1 Terminology 179 The common definitions used in this methodology are: 181 - TCP Throughput Test Device (TCP TTD), refers to compliant TCP 182 host that generates traffic and measures metrics as defined in 183 this methodology. i.e. a dedicated communications test instrument. 184 - Customer Provided Equipment (CPE), refers to customer owned 185 equipment (routers, switches, computers, etc.) 186 - Customer Edge (CE), refers to provider owned demarcation device. 187 - Provider Edge (PE), refers to provider's distribution equipment. 188 - Bottleneck Bandwidth (BB), lowest bandwidth along the complete 189 path. Bottleneck Bandwidth and Bandwidth are used synonymously 190 in this document. Most of the time the Bottleneck Bandwidth is 191 in the access portion of the wide area network (CE - PE). 192 - Provider (P), refers to provider core network equipment. 193 - Network Under Test (NUT), refers to the tested IP network path. 194 - Round Trip Time (RTT), is the elapsed time between the clocking in 195 of the first bit of a TCP segment sent and the receipt of the last 196 bit of the corresponding TCP Acknowledgment. 197 - Bandwidth Delay Product (BDP), refers to the product of a data 198 link's capacity (in bits per second) and its end-to-end delay 199 (in seconds). 201 Figure 1.1 Devices, Links and Paths 203 +----+ +----+ +----+ +----+ +---+ +---+ +----+ +----+ +----+ +----+ 204 | TCP|-| CPE|-| CE |--| PE |-| P |--| P |-| PE |--| CE |-| CPE|-| TCP| 205 | TTD| | | | |BB| | | | | | | |BB| | | | | TTD| 206 +----+ +----+ +----+ +----+ +---+ +---+ +----+ +----+ +----+ +----+ 207 <------------------------ NUT -------------------------> 208 R >-----------------------------------------------------------| 209 T | 210 T <-----------------------------------------------------------| 212 Note that the NUT may be built with of a variety of devices including 213 but not limited to, load balancers, proxy servers or WAN acceleration 214 appliances. The detailed topology of the NUT SHOULD be well known 215 when conducting the TCP Throughput tests, although this methodology 216 makes no attempt to characterize specific network architectures. 218 1.2 TCP Equilibrium 220 TCP connections have three (3) fundamental congestion window phases, 221 which are depicted in Figure 1.2. 223 1 - The Slow Start phase, which occurs at the beginning of a TCP 224 transmission or after a retransmission time out. 226 2 - The Congestion Avoidance phase, during which TCP ramps up to 227 establish the maximum achievable throughput. It is important to note 228 that retransmissions are a natural by-product of the TCP congestion 229 avoidance algorithm as it seeks to achieve maximum throughput. 231 3 - The Loss Recovery phase, which could include Fast Retransmit 232 (Tahoe) or Fast Recovery (Reno & New Reno). When packet loss occurs, 233 Congestion Avoidance phase transitions either to Fast Retransmission 234 or Fast Recovery depending upon the TCP implementation. If a Time-Out 235 occurs, TCP transitions back to the Slow Start phase. 237 Figure 1.2 TCP CWND Phases 239 /\ | 240 /\ |High ssthresh TCP CWND TCP 241 /\ |Loss Event * halving 3-Loss Recovery Equilibrium 242 /\ | * \ upon loss 243 /\ | * \ / \ Time-Out Adjusted 244 /\ | * \ / \ +--------+ * ssthresh 245 /\ | * \/ \ / Multiple| * 246 /\ | * 2-Congestion\ / Loss | * 247 /\ | * Avoidance \/ Event | * 248 TCP | * Half | * 249 Through- | * TCP CWND | * 1-Slow Start 250 put | * 1-Slow Start Min TCP CWND after T-O 251 +----------------------------------------------------------- 252 Time > > > > > > > > > > > > > > > > > > > > > > > > > > > 254 Note: ssthresh = Slow Start threshold. 256 A well tuned and managed IP network with appropriate TCP adjustments 257 in the IP hosts and applications should perform very close to the 258 BB when TCP is in the Equilibrium state. 260 This TCP methodology provides guidelines to measure the maximum 261 achievable TCP Throughput when TCP is in the Equilibrium state. 262 All maximum achievable TCP Throughputs specified in Section 3.3 are 263 with respect to this condition. 265 It is important to clarify the interaction between the sender's Send 266 Socket Buffer and the receiver's advertised TCP RWND Size. TCP test 267 programs such as iperf, ttcp, etc. allows the sender to control the 268 quantity of TCP Bytes transmitted and unacknowledged (in-flight), 269 commonly referred to as the Send Socket Buffer. This is done 270 independently of the TCP RWND Size advertised by the receiver. 272 2. Scope and Goals 274 Before defining the goals, it is important to clearly define the 275 areas that are out-of-scope. 277 - This methodology is not intended to predict the TCP Throughput 278 during the transient stages of a TCP connection, such as during the 279 slow start phase. 281 - This methodology is not intended to definitively benchmark TCP 282 implementations of one OS to another, although some users may find 283 value in conducting qualitative experiments. 285 - This methodology is not intended to provide detailed diagnosis 286 of problems within end-points or within the network itself as 287 related to non-optimal TCP performance, although results 288 interpretation for each test step may provide insights to potential 289 issues. 291 - This methodology does not propose to operate permanently with high 292 measurement loads. TCP performance and optimization within 293 operational networks MAY be captured and evaluated by using data 294 from the "TCP Extended Statistics MIB" [RFC4898]. 296 In contrast to the above exclusions, the primary goal is to define a 297 method to conduct a practical end-to-end assessment of sustained 298 TCP performance within a managed business class IP network. Another 299 key goal is to establish a set of "best practices" that a non-TCP 300 expert SHOULD apply when validating the ability of a managed IP 301 network to carry end-user TCP applications. 303 Specific goals are to: 305 - Provide a practical test approach that specifies tunable parameters 306 (such as MTU (Maximum Transmit Unit) and Socket Buffer sizes) and how 307 these affect the outcome of TCP performances over an IP network. 309 - Provide specific test conditions like link speed, RTT, MTU, Socket 310 Buffer sizes and achievable TCP Throughput when TCP is in the 311 Equilibrium state. For guideline purposes, provide examples of 312 test conditions and their maximum achievable TCP Throughput. 313 Section 1.2 provides specific details concerning the definition of 314 TCP Equilibrium within this methodology while Section 3 provides 315 specific test conditions with examples. 317 - Define three (3) basic metrics to compare the performance of TCP 318 connections under various network conditions. See Section 4. 320 - In test situations where the recommended procedure does not yield 321 the maximum achievable TCP Throughput, this methodology provides 322 some areas within the end host or the network that SHOULD be 323 considered for investigation. Although again, this methodology 324 is not intended to provide detailed diagnosis on these issues. 325 See Section 5.2. 327 3. Methodology 329 This methodology is intended for operational and managed IP networks. 330 A multitude of network architectures and topologies can be tested. 331 The diagram in Figure 1.1 is very general and is only there to 332 illustrate typical segmentation within end-user and network provider 333 domains. 335 Also, as stated earlier in Section 1, it is considered best practice 336 to verify the integrity of the network by conducting Layer 2/3 tests 337 such as [RFC2544] or other methods of network stress tests. 338 Although, it is important to mention here that [RFC2544] was never 339 meant to be used outside a lab environment. 341 It is not possible to make an accurate TCP Throughput measurement 342 when the network is dysfunctional. In particular, if the network is 343 exhibiting high packet loss and/or high jitter, then TCP Layer 344 Throughput testing will not be meaningful. As a guideline 5% packet 345 loss and/or 150 ms of jitter may be considered too high for an 346 accurate measurement. 348 TCP Throughput testing may require cooperation between the end-user 349 customer and the network provider. As an example, in an MPLS (Multi- 350 Protocol Label Switching) network architecture, the testing SHOULD be 351 conducted either on the CPE or on the CE device and not on the PE 352 (Provider Edge) router. 354 The following represents the sequential order of steps for this 355 testing methodology: 357 1 - Identify the Path MTU. Packetization Layer Path MTU Discovery 358 or PLPMTUD, [RFC4821], SHOULD be conducted. It is important to 359 identify the path MTU so that the TCP TTD is configured properly to 360 avoid fragmentation. 362 2 - Baseline Round Trip Time and Bandwidth. This step establishes the 363 inherent, non-congested Round Trip Time (RTT) and the Bottleneck 364 Bandwidth (BB) of the end-to-end network path. These measurements 365 are used to provide estimates of the TCP RWND and Send Socket Buffer 366 Sizes that SHOULD be used during subsequent test steps. 368 3 - TCP Connection Throughput Tests. With baseline measurements 369 of Round Trip Time and Bottleneck Bandwidth, single and multiple TCP 370 connection throughput tests SHOULD be conducted to baseline network 371 performances. 373 These three (3) steps are detailed in Sections 3.1 - 3.3. 375 Important to note are some of the key characteristics and 376 considerations for the TCP test instrument. The test host MAY be a 377 standard computer or a dedicated communications test instrument. 378 In both cases, it MUST be capable of emulating both a client and a 379 server. 381 The following criteria SHOULD be considered when selecting whether 382 the TCP test host can be a standard computer or has to be a dedicated 383 communications test instrument: 385 - TCP implementation used by the test host, OS version, i.e. LINUX OS 386 kernel using TCP New Reno, TCP options supported, etc. These will 387 obviously be more important when using dedicated communications test 388 instruments where the TCP implementation may be customized or tuned 389 to run in higher performance hardware. When a compliant TCP TTD is 390 used, the TCP implementation SHOULD be identified in the test 391 results. The compliant TCP TTD SHOULD be usable for complete 392 end-to-end testing through network security elements and SHOULD also 393 be usable for testing network sections. 395 - More important, the TCP test host MUST be capable to generate 396 and receive stateful TCP test traffic at the full BB of the NUT. 397 Stateful TCP test traffic means that the test host MUST fully 398 implement a TCP/IP stack; this is generally a comment aimed at 399 dedicated communications test equipments which sometimes "blast" 400 packets with TCP headers. As a general rule of thumb, testing TCP 401 Throughput at rates greater than 100 Mbps may require high 402 performance server hardware or dedicated hardware based test tools. 404 - A compliant TCP Throughput Test Device MUST allow adjusting both 405 Send and Receive Socket Buffer sizes. The Socket Buffers MUST be 406 large enough to fill the BDP. 408 - Measuring RTT and retransmissions per connection will generally 409 require a dedicated communications test instrument. In the absence of 410 dedicated hardware based test tools, these measurements may need to 411 be conducted with packet capture tools, i.e. conduct TCP Throughput 412 tests and analyze RTT and retransmissions in packet captures. 413 Another option MAY be to use "TCP Extended Statistics MIB" per 414 [RFC4898]. 416 - The [RFC4821] PLPMTUD test SHOULD be conducted with a dedicated 417 tester which exposes the ability to run the PLPMTUD algorithm 418 independently from the OS stack. 420 3.1. Path MTU 422 TCP implementations should use Path MTU Discovery techniques (PMTUD). 423 PMTUD relies on ICMP 'need to frag' messages to learn the path MTU. 424 When a device has a packet to send which has the Don't Fragment (DF) 425 bit in the IP header set and the packet is larger than the (MTU) of 426 the next hop, the packet is dropped and the device sends an ICMP 427 'need to frag' message back to the host that originated the packet. 428 The ICMP 'need to frag' message includes the next hop MTU which PMTUD 429 uses to adjust itself. Unfortunately, because many network managers 430 completely disable ICMP, this technique does not always prove 431 reliable. 433 Packetization Layer Path MTU Discovery or PLPMTUD [RFC4821] MUST then 434 be conducted to verify the network path MTU. PLPMTUD can be used 435 with or without ICMP. [RFC4821] specifies search_high and search_low 436 parameters for the MTU and we recommend to use those. The goal is to 437 avoid fragmentation during all subsequent tests. 439 3.2. Round Trip Time (RTT) and Bottleneck Bandwidth (BB) 441 Before stateful TCP testing can begin, it is important to determine 442 the baseline RTT (i.e. non-congested inherent delay) and BB of the 443 end-to-end network to be tested. These measurements are used to 444 calculate the BDP and to provide estimates of the TCP RWND and 445 Send Socket Buffer Sizes that SHOULD be used in subsequent test 446 steps. 448 3.2.1 Measuring RTT 450 As previously defined in Section 1.1, RTT is the elapsed time 451 between the clocking in of the first bit of a TCP segment sent 452 and the receipt of the last bit of the corresponding TCP 453 Acknowledgment. 455 The RTT SHOULD be baselined during off-peak hours in order to obtain 456 a reliable figure of the inherent network latency. Otherwise, 457 additional delay caused by network buffering can occur. Also, when 458 sampling RTT values over a given test interval, the minimum 459 measured value SHOULD be used as the baseline RTT. This will most 460 closely estimate the real inherent RTT. This value is also used to 461 determine the Buffer Delay Percentage metric defined in Section 4.3. 463 The following list is not meant to be exhaustive, although it 464 summarizes some of the most common ways to determine Round Trip Time. 465 The desired measurement precision (i.e. ms versus us) may dictate 466 whether the RTT measurement can be achieved with ICMP pings or by a 467 dedicated communications test instrument with precision timers. The 468 objective in this section is to list several techniques in order of 469 decreasing accuracy. 471 - Use test equipment on each end of the network, "looping" the 472 far-end tester so that a packet stream can be measured back and forth 473 from end-to-end. This RTT measurement may be compatible with delay 474 measurement protocols specified in [RFC5357]. 476 - Conduct packet captures of TCP test sessions using "iperf" or FTP, 477 or other TCP test applications. By running multiple experiments, 478 packet captures can then be analyzed to estimate RTT. It is 479 important to note that results based upon the SYN -> SYN-ACK at the 480 beginning of TCP sessions SHOULD be avoided since Firewalls might 481 slow down 3 way handshakes. Also, at the senders side, Ostermann's 482 LINUX TCPTRACE utility with -l -r arguments can be used to extract 483 the RTT results directly from the packet captures. 485 - Obtain RTT statistics available from MIBs defined in [RFC4898]. 487 - ICMP pings may also be adequate to provide Round Trip Time 488 estimates, provided that the packet size is factored into the 489 estimates (i.e. pings with different packet sizes might be required). 490 Some limitations with ICMP Ping may include ms resolution and 491 whether the network elements are responding to pings or not. Also, 492 ICMP is often rate-limited or segregated into different buffer 493 queues. ICMP might not work if QoS (Quality of Service) 494 reclassification is done at any hop. ICMP is not as reliable and 495 accurate as in-band measurements. 497 3.2.2 Measuring BB 499 Before any TCP Throughput test can be conducted, bandwidth 500 measurement tests SHOULD be run with stateless IP streams (i.e. not 501 stateful TCP) in order to determine the BB of the NUT. 502 These measurements SHOULD be conducted in both directions, 503 especially in asymmetrical access networks (e.g. ADSL access). These 504 tests SHOULD be performed at various intervals throughout a business 505 day or even across a week. 507 Testing at various time intervals would provide a better 508 characterization of TCP throughput and better diagnosis insight (for 509 cases where there are TCP performance issues). The bandwidth tests 510 SHOULD produce logged outputs of the achieved bandwidths across the 511 complete test duration. 513 There are many well established techniques available to provide 514 estimated measures of bandwidth over a network. It is a common 515 practice for network providers to conduct Layer 2/3 bandwidth 516 capacity tests using [RFC2544], although it is understood that 517 [RFC2544] was never meant to be used outside a lab environment. 518 These bandwidth measurements SHOULD use network capacity 519 techniques as defined in [RFC5136]. 521 3.3. Measuring TCP Throughput 523 This methodology specifically defines TCP Throughput measurement 524 techniques to verify maximum achievable TCP performance in a managed 525 business class IP network. 527 With baseline measurements of RTT and BB from Section 3.2, a series 528 of single and / or multiple TCP connection throughput tests SHOULD 529 be conducted. 531 The number of trials and single versus multiple TCP connections 532 choice will be based on the intention of the test. A single TCP 533 connection test might be enough to measure the achievable throughput 534 of a Metro Ethernet connectivity. Although, it is important to note 535 that various traffic management techniques can be used in an IP 536 network and that some of those can only be tested with multiple 537 connections. As an example, multiple TCP sessions might be required 538 to detect traffic shaping versus policing. Multiple sessions might 539 also be needed to measure Active Queue Management performances. 540 However, traffic management testing is not within the scope of this 541 test methodology. 543 In all circumstances, it is RECOMMENDED to run the tests in each 544 direction independently first and then to run in both directions 545 simultaneously. It is also RECOMMENDED to run the tests at 546 different times of day. 548 In each case, the TCP Transfer Time Ratio, the TCP Efficiency 549 Percentage, and the Buffer Delay Percentage MUST be measured in 550 each direction. These 3 metrics are defined in Section 4. 552 3.3.1 Minimum TCP RWND 554 The TCP TTD MUST allow the Send Socket Buffer and Receive Window 555 sizes to be set higher than the BDP, otherwise TCP performance will 556 be limited. In the business customer environment, these settings are 557 not generally adjustable by the average user. These settings are 558 either hard coded in the application or configured within the OS as 559 part of a corporate image. In many cases, the user's host Send 560 Socket Buffer and Receive Window size settings are not optimal. 562 This section provides derivations of BDPs under various network 563 conditions. It also provides examples of achievable TCP Throughput 564 with various TCP RWND sizes. This provides important guidelines 565 showing what can be achieved with settings higher than the BDP, 566 versus what would be achieved in a variety of real world conditions. 568 The minimum required TCP RWND Size can be calculated from the 569 Bandwidth Delay Product (BDP), which is: 571 BDP (bits) = RTT (sec) x BB (bps) 572 Note that the RTT is being used as the "Delay" variable for the BDP. 574 Then, by dividing the BDP by 8, we obtain the minimum required TCP 575 RWND Size in Bytes. For optimal results, the Send Socket Buffer 576 MUST be adjusted to the same value at each end of the network. 578 Minimum required TCP RWND = BDP / 8 580 As an example on a T3 link with 25 ms RTT, the BDP would equal 581 ~1,105,000 bits and the minimum required TCP RWND would be ~138 KB. 583 Note that separate calculations are REQUIRED on asymmetrical paths. 584 An asymmetrical path example would be a 90 ms RTT ADSL line with 585 5Mbps downstream and 640Kbps upstream. The downstream BDP would equal 586 ~450,000 bits while the upstream one would be only ~57,600 bits. 588 The following table provides some representative network Link Speeds, 589 RTT, BDP, and their associated minimum required TCP RWND Sizes. 591 Table 3.3.1: Link Speed, RTT, calculated BDP & min. TCP RWND 593 Link Minimum required 594 Speed* RTT BDP TCP RWND 595 (Mbps) (ms) (bits) (KBytes) 596 --------------------------------------------------------------------- 597 1.536 20.00 30,720 3.84 598 1.536 50.00 76,800 9.60 599 1.536 100.00 153,600 19.20 600 44.210 10.00 442,100 55.26 601 44.210 15.00 663,150 82.89 602 44.210 25.00 1,105,250 138.16 603 100.000 1.00 100,000 12.50 604 100.000 2.00 200,000 25.00 605 100.000 5.00 500,000 62.50 606 1,000.000 0.10 100,000 12.50 607 1,000.000 0.50 500,000 62.50 608 1,000.000 1.00 1,000,000 125.00 609 10,000.000 0.05 500,000 62.50 610 10,000.000 0.30 3,000,000 375.00 612 * Note that link speed is the BB for the NUT 613 In the above table, the following serial link speeds are used: 614 - T1 = 1.536 Mbps (for a B8ZS line encoding facility) 615 - T3 = 44.21 Mbps (for a C-Bit Framing facility) 617 The previous table illustrates the minimum required TCP RWND. 618 If a smaller TCP RWND Size is used, then the TCP Throughput 619 can not be optimal. To calculate the TCP Throughput, the following 620 formula is used: TCP Throughput = TCP RWND X 8 / RTT 622 An example could be a 100 Mbps IP path with 5 ms RTT and a TCP RWND 623 of 16KB, then: 625 TCP Throughput = 16 KBytes X 8 bits / 5 ms. 626 TCP Throughput = 128,000 bits / 0.005 sec. 627 TCP Throughput = 25.6 Mbps. 629 Another example for a T3 using the same calculation formula is 630 illustrated in Figure 3.3.1a: 632 TCP Throughput = 16 KBytes X 8 bits / 10 ms. 633 TCP Throughput = 128,000 bits / 0.01 sec. 634 TCP Throughput = 12.8 Mbps. * 636 When the TCP RWND Size exceeds the BDP (T3 link and 64 KBytes TCP 637 RWND on a 10 ms RTT path), the maximum frames per second limit of 638 3664 is reached and then the formula is: 640 TCP Throughput = Max FPS X (MTU - 40) X 8. 641 TCP Throughput = 3664 FPS X 1460 Bytes X 8 bits. 642 TCP Throughput = 42.8 Mbps. ** 644 The following diagram compares achievable TCP Throughputs on a T3 645 with Send Socket Buffer & TCP RWND Sizes of 16KB vs. 64KB. 647 Figure 3.3.1a TCP Throughputs on a T3 at different RTTs 649 45| 650 | _______**42.8 651 40| |64KB | 652 TCP | | | 653 Throughput 35| | | 654 in Mbps | | | +-----+34.1 655 30| | | |64KB | 656 | | | | | 657 25| | | | | 658 | | | | | 659 20| | | | | _______20.5 660 | | | | | |64KB | 661 15| | | | | | | 662 |*12.8+-----| | | | | | 663 10| |16KB | | | | | | 664 | | | |8.5 +-----| | | | 665 5| | | | |16KB | |5.1 +-----| | 666 |_____|_____|_____|____|_____|_____|____|16KB |_____|_____ 667 10 15 25 668 RTT in milliseconds 670 The following diagram shows the achievable TCP Throughput on a 25 ms 671 T3 when Send Socket Buffer & TCP RWND Sizes are increased. 673 Figure 3.3.1b TCP Throughputs on a T3 with different TCP RWND 675 45| 676 | 677 40| +-----+40.9 678 TCP | | | 679 Throughput 35| | | 680 in Mbps | | | 681 30| | | 682 | | | 683 25| | | 684 | | | 685 20| +-----+20.5 | | 686 | | | | | 687 15| | | | | 688 | | | | | 689 10| +-----+10.2 | | | | 690 | | | | | | | 691 5| +-----+5.1 | | | | | | 692 |_____|_____|______|_____|______|_____|_______|_____|_____ 693 16 32 64 128* 694 TCP RWND Size in KBytes 696 * Note that 128KB requires [RFC1323] TCP Window scaling option. 698 4. TCP Metrics 700 This methodology focuses on a TCP Throughput and provides 3 basic 701 metrics that can be used for better understanding of the results. 702 It is recognized that the complexity and unpredictability of TCP 703 makes it very difficult to develop a complete set of metrics that 704 accounts for the myriad of variables (i.e. RTT variations, loss 705 conditions, TCP implementations, etc.). However, these 3 metrics 706 facilitate TCP Throughput comparisons under varying network 707 conditions and host buffer size / RWND settings. 709 4.1 Transfer Time Ratio 711 The first metric is the TCP Transfer Time Ratio, which is simply the 712 ratio between the Actual versus the Ideal TCP Transfer Times. 714 The Actual TCP Transfer Time, is simply the time it takes to transfer 715 a block of data across TCP connection(s). 717 The Ideal TCP Transfer Time is the predicted time for which a block 718 of data SHOULD transfer across TCP connection(s) considering the BB 719 of the NUT. 721 Actual TCP Transfer Time 722 TCP Transfer Time Ratio = ------------------------- 723 Ideal TCP Transfer Time 725 The Ideal TCP Transfer Time is derived from the Maximum Achievable 726 TCP Throughput, which is related to the BB and Layer 1/2/3/4 727 overheads associated with the network path. The following sections 728 provide derivations for the Maximum Achievable TCP Throughput and 729 example calculations for the TCP Transfer Time Ratio. 731 4.1.1 Maximum Achievable TCP Throughput calculation 733 This section provides formulas to calculate the Maximum Achievable 734 TCP Throughput with examples for T3 (44.21 Mbps) and Ethernet. 736 All calculations are based on IP version 4 with TCP/IP headers of 737 20 Bytes each (20 for TCP + 20 for IP) within an MTU of 1500 Bytes. 739 First, the maximum achievable Layer 2 throughput of a T3 Interface 740 is limited by the maximum quantity of Frames Per Second (FPS) 741 permitted by the actual physical layer (Layer 1) speed. 743 The calculation formula is: 744 FPS = T3 Physical Speed / ((MTU + PPP + Flags + CRC16) X 8) 745 FPS = (44.21Mbps /((1500 Bytes + 4 Bytes + 2 Bytes + 2 Bytes) X 8 ))) 746 FPS = (44.21Mbps /(1508 Bytes X 8)) 747 FPS = 44.21Mbps / 12064 bits 748 FPS = 3664 750 Then, to obtain the Maximum Achievable TCP Throughput (Layer 4), we 751 simply use: (MTU - 40) in Bytes X 8 bits X max FPS. 753 For a T3, the maximum TCP Throughput = 1460 Bytes X 8 bits X 3664 FPS 754 Maximum TCP Throughput = 11680 bits X 3664 FPS 755 Maximum TCP Throughput = 42.8 Mbps. 757 On Ethernet, the maximum achievable Layer 2 throughput is limited by 758 the maximum Frames Per Second permitted by the IEEE802.3 standard. 760 The maximum FPS for 100 Mbps Ethernet is 8127 and the calculation is: 761 FPS = (100Mbps /(1538 Bytes X 8 bits)) 763 The maximum FPS for GigE is 81274 and the calculation formula is: 764 FPS = (1Gbps /(1538 Bytes X 8 bits)) 766 The maximum FPS for 10GigE is 812743 and the calculation formula is: 767 FPS = (10Gbps /(1538 Bytes X 8 bits)) 769 The 1538 Bytes equates to: 771 MTU + Ethernet + CRC32 + IFG + Preamble + SFD 772 (IFG = Inter-Frame Gap and SFD = Start of Frame Delimiter) 773 Where MTU is 1500 Bytes, Ethernet is 14 Bytes, CRC32 is 4 Bytes, 774 IFG is 12 Bytes, Preamble is 7 Bytes and SFD is 1 Byte. 776 Then, to obtain the Maximum Achievable TCP Throughput (Layer 4), we 777 simply use: (MTU - 40) in Bytes X 8 bits X max FPS. 778 For a 100Mbps, the max TCP Throughput = 1460Bytes X 8 bits X 8127 FPS 779 Maximum TCP Throughput = 11680 bits X 8127 FPS 780 Maximum TCP Throughput = 94.9 Mbps. 782 It is important to note that better results could be obtained with 783 jumbo frames on Gigabit and 10 Gigabit Ethernet interfaces. 785 4.1.2 TCP Transfer Time and Transfer Time Ratio calculation 787 The following table illustrates the Ideal TCP Transfer time of a 788 single TCP connection when its TCP RWND and Send Socket Buffer Sizes 789 equals or exceeds the BDP. 791 Table 4.1.1: Link Speed, RTT, BDP, TCP Throughput, and 792 Ideal TCP Transfer time for a 100 MB File 794 Link Maximum Ideal TCP 795 Speed BDP Achievable TCP Transfer time 796 (Mbps) RTT (ms) (KBytes) Throughput(Mbps) (seconds)* 797 -------------------------------------------------------------------- 798 1.536 50.00 9.6 1.4 571.0 799 44.210 25.00 138.2 42.8 18.0 800 100.000 2.00 25.0 94.9 9.0 801 1,000.000 1.00 125.0 949.2 1.0 802 10,000.000 0.05 62.5 9,492.0 0.1 804 * Transfer times are rounded for simplicity. 806 For a 100MB file (100 x 8 = 800 Mbits), the Ideal TCP Transfer Time 807 is derived as follows: 809 800 Mbits 810 Ideal TCP Transfer Time = ----------------------------------- 811 Maximum Achievable TCP Throughput 813 To illustrate the TCP Transfer Time Ratio, an example would be the 814 bulk transfer of 100 MB over 5 simultaneous TCP connections (each 815 connection transferring 100 MB). In this example, the Ethernet 816 service provides a Committed Access Rate (CAR) of 500 Mbps. Each 817 connection may achieve different throughputs during a test and the 818 overall throughput rate is not always easy to determine (especially 819 as the number of connections increases). 821 The ideal TCP Transfer Time would be ~8 seconds, but in this example, 822 the actual TCP Transfer Time was 12 seconds. The TCP Transfer Time 823 Ratio would then be 12/8 = 1.5, which indicates that the transfer 824 across all connections took 1.5 times longer than the ideal. 826 4.2 TCP Efficiency 828 The second metric represents the percentage of Bytes that were not 829 retransmitted. 831 Transmitted Bytes - Retransmitted Bytes 832 TCP Efficiency % = --------------------------------------- X 100 833 Transmitted Bytes 835 Transmitted Bytes are the total number of TCP Bytes to be transmitted 836 including the original and the retransmitted Bytes. 838 4.2.1 TCP Efficiency Percentage calculation 840 As an example, if 100,000 Bytes were sent and 2,000 had to be 841 retransmitted, the TCP Efficiency Percentage would be calculated as: 843 102,000 - 2,000 844 TCP Efficiency % = ----------------- x 100 = 98.03% 845 102,000 847 Note that the Retransmitted Bytes may have occurred more than once, 848 if so, then these multiple retransmissions are added to the 849 Retransmitted Bytes and to the Transmitted Bytes counts. 851 4.3 Buffer Delay 853 The third metric is the Buffer Delay Percentage, which represents 854 the increase in RTT during a TCP Throughput test versus the inherent 855 or baseline RTT. The baseline RTT is the Round Trip Time inherent to 856 the network path under non-congested conditions as defined in Section 857 3.2.1. The average RTT is derived from the total of all measured 858 RTTs during the actual test at every second divided by the test 859 duration in seconds. 861 Total RTTs during transfer 862 Average RTT during transfer = ----------------------------- 863 Transfer duration in seconds 865 Average RTT during Transfer - Baseline RTT 866 Buffer Delay % = ------------------------------------------ X 100 867 Baseline RTT 869 4.3.1 Buffer Delay calculation 871 As an example, consider a network path with a baseline RTT of 25 ms. 872 During the course of a TCP transfer, the average RTT across 873 the entire transfer increases to 32 ms. Then, the Buffer Delay 874 Percentage would be calculated as: 876 32 - 25 877 Buffer Delay % = ------- x 100 = 28% 878 25 880 Note that the TCP Transfer Time Ratio, TCP Efficiency Percentage, and 881 the Buffer Delay Percentage MUST all be measured during each 882 throughput test. Poor TCP Transfer Time Ratio (i.e. TCP Transfer 883 Time greater than the Ideal TCP Transfer Time) may be diagnosed by 884 correlating with sub-optimal TCP Efficiency Percentage and/or Buffer 885 Delay Percentage metrics. 887 5. Conducting TCP Throughput Tests 889 Several TCP tools are currently used in the network world and one of 890 the most common is "iperf". With this tool, hosts are installed at 891 each end of the network path; one acts as client and the other as 892 a server. The Send Socket Buffer and the TCP RWND Sizes of both 893 client and server can be manually set. The achieved throughput can 894 then be measured, either uni-directionally or bi-directionally. For 895 higher BDP situations in lossy networks (Long Fat Networks (LFNs) or 896 satellite links, etc.), TCP options such as Selective Acknowledgment 897 SHOULD become part of the window size / throughput characterization. 899 Host hardware performance must be well understood before conducting 900 the tests described in the following sections. A dedicated 901 communications test instrument will generally be REQUIRED, especially 902 for line rates of GigE and 10 GigE. A compliant TCP TTD SHOULD 903 provide a warning message when the expected test throughput will 904 exceed the subscribed customer SLA. If the throughput test is 905 expected to exceed the subscribed customer SLA, then the test 906 SHOULD be coordinated with the network provider. 908 The TCP Throughput test SHOULD be run over a long enough duration 909 to properly exercise network buffers (i.e. greater than 30 seconds) 910 and SHOULD also characterize performance at different times of day. 912 5.1 Single versus Multiple TCP Connections 914 The decision whether to conduct single or multiple TCP connection 915 tests depends upon the size of the BDP in relation to the TCP RWND 916 configured in the end-user environment. For example, if the BDP for 917 a Long Fat Network (LFN) turns out to be 2MB, then it is probably 918 more realistic to test this network path with multiple connections. 919 Assuming typical host TCP RWND Sizes of 64 KB (i.e. Windows XP), 920 using 32 TCP connections would emulate a small office scenario. 922 The following table is provided to illustrate the relationship 923 between the TCP RWND and the number of TCP connections required to 924 fill the available capacity of a given BDP. For this example, the 925 network bandwidth is 500 Mbps and the RTT is 5 ms, then the BDP 926 equates to 312.5 KBytes. 928 Table 5.1 Number of TCP connections versus TCP RWND 930 Number of TCP Connections 931 TCP RWND to fill available bandwidth 932 ------------------------------------- 933 16KB 20 934 32KB 10 935 64KB 5 936 128KB 3 938 The TCP Transfer Time Ratio metric is useful when conducting multiple 939 connection tests. Each connection SHOULD be configured to transfer 940 payloads of the same size (i.e. 100 MB), then the TCP Transfer Time 941 Ratio provides a simple metric to verify the actual versus expected 942 results. 944 Note that the TCP Transfer Time is the time required for each 945 connection to complete the transfer of the predetermined payload 946 size. From the previous table, the 64KB window is considered. Each 947 of the 5 TCP connections would be configured to transfer 100MB, and 948 each one should obtain a maximum of 100 Mbps. So for this example, 949 the 100MB payload should be transferred across the connections in 950 approximately 8 seconds (which would be the Ideal TCP Transfer Time 951 under these conditions). 953 Additionally, the TCP Efficiency Percentage metric MUST be computed 954 for each connection as defined in Section 4.2. 956 5.2 Results Interpretation 958 At the end, a TCP Throughput Test Device (TCP TTD) SHOULD generate a 959 report with the calculated BDP and a set of Window Size experiments. 960 Window Size refers to the minimum of the Send Socket Buffer and TCP 961 RWND. The report SHOULD include TCP Throughput results for each TCP 962 Window Size tested. The goal is to provide clear achievable versus 963 actual TCP Throughputs results with respect to the TCP Window Size 964 when no fragmentation occurs. The report SHOULD also include the 965 results for the 3 metrics defined in Section 4. The goal is to 966 provide a clear relationship between these 3 metrics and user 967 experience. As an example, for the same results in regards with 968 Transfer Time Ratio, a better TCP Efficiency could be obtained at the 969 cost of higher Buffer Delays. 971 For cases where the test results are not equal to the ideal values, 972 some possible causes are: 974 - Network congestion causing packet loss which may be inferred from 975 a poor TCP Efficiency % (i.e., higher TCP Efficiency % = less packet 976 loss) 978 - Network congestion causing an increase in RTT which may be inferred 979 from the Buffer Delay Percentage (i.e., 0% = no increase in RTT over 980 baseline) 981 - Intermediate network devices which actively regenerate the TCP 982 connection and can alter TCP RWND Size, MTU, etc. 984 - Rate limiting by policing instead of shaping. 986 - Maximum TCP Buffer space. All operating systems have a global 987 mechanism to limit the quantity of system memory to be used by TCP 988 connections. On some systems, each connection is subject to a memory 989 limit that is applied to the total memory used for input data, output 990 data and controls. On other systems, there are separate limits for 991 input and output buffer spaces per connection. Client/server IP 992 hosts might be configured with Maximum Buffer Space limits that are 993 far too small for high performance networks. 995 - Socket Buffer Sizes. Most operating systems support separate per 996 connection send and receive buffer limits that can be adjusted as 997 long as they stay within the maximum memory limits. These socket 998 buffers MUST be large enough to hold a full BDP of TCP Bytes plus 999 some overhead. There are several methods that can be used to adjust 1000 socket buffer sizes, but TCP Auto-Tuning automatically adjusts these 1001 as needed to optimally balance TCP performance and memory usage. 1003 It is important to note that Auto-Tuning is enabled by default in 1004 LINUX since the kernel release 2.6.6 and in UNIX since FreeBSD 7.0. 1005 It is also enabled by default in Windows since Vista and in MAC since 1006 OS X version 10.5 (leopard). Over buffering can cause some 1007 applications to behave poorly, typically causing sluggish interactive 1008 response and risk running the system out of memory. Large default 1009 socket buffers have to be considered carefully on multi-user systems. 1011 - TCP Window Scale Option, [RFC1323]. This option enables TCP to 1012 support large BDP paths. It provides a scale factor which is 1013 required for TCP to support window sizes larger than 64KB. Most 1014 systems automatically request WSCALE under some conditions, such as 1015 when the receive socket buffer is larger than 64KB or when the other 1016 end of the TCP connection requests it first. WSCALE can only be 1017 negotiated during the 3 way handshake. If either end fails to 1018 request WSCALE or requests an insufficient value, it cannot be 1019 renegotiated. Different systems use different algorithms to select 1020 WSCALE, but it is very important to have large enough buffer 1021 sizes. Note that under these constraints, a client application 1022 wishing to send data at high rates may need to set its own receive 1023 buffer to something larger than 64K Bytes before it opens the 1024 connection to ensure that the server properly negotiates WSCALE. 1025 A system administrator might have to explicitly enable [RFC1323] 1026 extensions. Otherwise, the client/server IP host would not support 1027 TCP window sizes (BDP) larger than 64KB. Most of the time, 1028 performance gains will be obtained by enabling this option in LFNs. 1030 - TCP Timestamps Option, [RFC1323]. This feature provides better 1031 measurements of the Round Trip Time and protects TCP from data 1032 corruption that might occur if packets are delivered so late that the 1033 sequence numbers wrap before they are delivered. Wrapped sequence 1034 numbers do not pose a serious risk below 100 Mbps, but the risk 1035 increases at higher data rates. Most of the time, performance gains 1036 will be obtained by enabling this option in Gigabit bandwidth 1037 networks. 1039 - TCP Selective Acknowledgments Option (SACK), [RFC2018]. This allows 1040 a TCP receiver to inform the sender about exactly which data segment 1041 is missing and needs to be retransmitted. Without SACK, TCP has to 1042 estimate which data segment is missing, which works just fine if all 1043 losses are isolated (i.e. only one loss in any given round trip). 1044 Without SACK, TCP takes a very long time to recover after multiple 1045 and consecutive losses. SACK is now supported by most operating 1046 systems, but it may have to be explicitly enabled by the system 1047 administrator. In networks with unknown load and error patterns, TCP 1048 SACK will improve throughput performances. On the other hand, 1049 security appliances vendors might have implemented TCP randomization 1050 without considering TCP SACK and under such circumstances, SACK might 1051 need to be disabled in the client/server IP hosts until the vendor 1052 corrects the issue. Also, poorly implemented SACK algorithms might 1053 cause extreme CPU loads and might need to be disabled. 1055 - Path MTU. The client/server IP host system SHOULD use the largest 1056 possible MTU for the path. This may require enabling Path MTU 1057 Discovery [RFC1191] & [RFC4821]. Since [RFC1191] is flawed, it is 1058 sometimes not enabled by default and may need to be explicitly 1059 enabled by the system administrator. [RFC4821] describes a new, more 1060 robust algorithm for MTU discovery and ICMP black hole recovery. 1062 - TOE (TCP Offload Engine). Some recent Network Interface Cards (NIC) 1063 are equipped with drivers that can do part or all of the TCP/IP 1064 protocol processing. TOE implementations require additional work 1065 (i.e. hardware-specific socket manipulation) to set up and tear down 1066 connections. Because TOE NICs configuration parameters are vendor 1067 specific and not necessarily RFC-compliant, they are poorly 1068 integrated with UNIX & LINUX. Occasionally, TOE might need to be 1069 disabled in a server because its NIC does not have enough memory 1070 resources to buffer thousands of connections. 1072 Note that both ends of a TCP connection MUST be properly tuned. 1074 6. Security Considerations 1076 Measuring TCP network performance raises security concerns. Metrics 1077 produced within this framework may create security issues. 1079 6.1 Denial of Service Attacks 1081 TCP network performance metrics, as defined in this document attempts 1082 to fill the NUT with a stateful connection. However, since the test 1083 MAY use stateless IP streams as specified in Section 3.2.2, it might 1084 appear to network operators as a Denial Of Service attack. Thus, as 1085 mentioned at the beginning of section 3, TCP Throughput testing may 1086 require cooperation between the end-user customer and the network 1087 provider. 1089 6.2 User data confidentiality 1091 Metrics within this framework generate packets from a sample, rather 1092 than taking samples based on user data. Thus, our framework does not 1093 threaten user data confidentiality. 1095 6.3 Interference with metrics 1097 The security considerations that apply to any active measurement of 1098 live networks are relevant here as well. See [RFC4656] and 1099 [RFC5357]. 1101 7. IANA Considerations 1103 This document does not REQUIRE an IANA registration for ports 1104 dedicated to the TCP testing described in this document. 1106 8. Acknowledgments 1108 Thanks to Lars Eggert, Al Morton, Matt Mathis, Matt Zekauskas, 1109 Yaakov Stein, and Loki Jorgenson for many good comments and for 1110 pointing us to great sources of information pertaining to past works 1111 in the TCP capacity area. 1113 9. References 1115 9.1 Normative References 1117 [RFC1191] Mogul, A., Deering, S., "Path MTU Discovery", 1990 1119 [RFC1323] Jacobson, V., Braden, R., Borman D., "TCP Extensions for 1120 High Performance", May 1992 1122 [RFC2018] Mathis, M., Mahdavi, J., Floyd, S., Romanow, A., "TCP 1123 Selective Acknowledgment Options", 1996 1125 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 1126 Requirement Levels", BCP 14, RFC 2119, March 1997. 1128 [RFC2544] Bradner, S., McQuaid, J., "Benchmarking Methodology for 1129 Network Interconnect Devices", RFC 2544, June 1999 1131 [RFC4656] Shalunov, S., Teitelbaum, B., Karp, A., Boote, J., and M. 1132 Zekauskas, "A One-way Active Measurement Protocol 1133 (OWAMP)", RFC 4656, September 2006. 1135 [RFC4821] Mathis, M., Heffner, J., "Packetization Layer Path MTU 1136 Discovery", RFC 4821, June 2007 1138 [RFC4898] Mathis, M., Heffner, J., Raghunarayan, R., "TCP Extended 1139 Statistics MIB", May 2007 1141 [RFC5136] Chimento P., Ishac, J., "Defining Network Capacity", 1142 February 2008 1144 [RFC5357] Hedayat, K., Krzanowski, R., Morton, A., Yum, K., Babiarz, 1145 J., "A Two-Way Active Measurement Protocol (TWAMP)", 1146 RFC 5357, October 2008 1148 draft-ietf-ippm-btc-cap-00.txt Allman, M., "A Bulk 1149 Transfer Capacity Methodology for Cooperating Hosts", 1150 August 2001 1152 9.2. Informative References 1153 Authors' Addresses 1155 Barry Constantine 1156 JDSU, Test and Measurement Division 1157 One Milesone Center Court 1158 Germantown, MD 20876-7100 1159 USA 1161 Phone: +1 240 404 2227 1162 barry.constantine@jdsu.com 1164 Gilles Forget 1165 Independent Consultant to Bell Canada. 1166 308, rue de Monaco, St-Eustache 1167 Qc. CANADA, Postal Code: J7P-4T5 1169 Phone: (514) 895-8212 1170 gilles.forget@sympatico.ca 1172 Ruediger Geib 1173 Heinrich-Hertz-Strasse (Number: 3-7) 1174 Darmstadt, Germany, 64295 1176 Phone: +49 6151 6282747 1177 Ruediger.Geib@telekom.de 1179 Reinhard Schrage 1180 Schrage Consulting 1181 Phone: +49 (0) 5137 909540 1182 reinhard@schrageconsult.com