idnits 2.17.1 draft-morton-ippm-2679-bis-06.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 : ---------------------------------------------------------------------------- -- The draft header indicates that this document obsoletes RFC2679, but the abstract doesn't seem to directly say this. It does mention RFC2679 though, so this could be OK. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (October 6, 2014) is 3488 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) == Unused Reference: 'RFC2026' is defined on line 1030, but no explicit reference was found in the text == Unused Reference: 'RFC4656' is defined on line 1053, but no explicit reference was found in the text == Unused Reference: 'RFC5357' is defined on line 1057, but no explicit reference was found in the text == Unused Reference: 'RFC5657' is defined on line 1061, but no explicit reference was found in the text == Unused Reference: 'RFC5835' is defined on line 1065, but no explicit reference was found in the text == Unused Reference: 'RFC5905' is defined on line 1068, but no explicit reference was found in the text == Unused Reference: 'RFC6049' is defined on line 1072, but no explicit reference was found in the text == Unused Reference: 'ADK' is defined on line 1090, but no explicit reference was found in the text == Unused Reference: 'I-D.ietf-ippm-testplan-rfc2679' is defined on line 1095, but no explicit reference was found in the text == Unused Reference: 'RFC3931' is defined on line 1101, but no explicit reference was found in the text ** Downref: Normative reference to an Informational RFC: RFC 2330 ** Obsolete normative reference: RFC 2679 (Obsoleted by RFC 7679) ** Obsolete normative reference: RFC 2680 (Obsoleted by RFC 7680) ** Downref: Normative reference to an Informational RFC: RFC 5835 ** Downref: Normative reference to an Informational RFC: RFC 6703 ** Obsolete normative reference: RFC 7321 (Obsoleted by RFC 8221) Summary: 6 errors (**), 0 flaws (~~), 11 warnings (==), 2 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Network Working Group G. Almes 3 Internet-Draft Texas A&M 4 Obsoletes: 2679 (if approved) S. Kalidindi 5 Intended status: Standards Track Ixia 6 Expires: April 9, 2015 M. Zekauskas 7 Internet2 8 A. Morton, Ed. 9 AT&T Labs 10 October 6, 2014 12 A One-Way Delay Metric for IPPM 13 draft-morton-ippm-2679-bis-06 15 Abstract 17 This memo (RFC 2679 bis) defines a metric for one-way delay of 18 packets across Internet paths. It builds on notions introduced and 19 discussed in the IPPM Framework document, RFC 2330; the reader is 20 assumed to be familiar with that document. 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 April 9, 2015. 45 Copyright Notice 47 Copyright (c) 2014 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. RFC 2679 bis . . . . . . . . . . . . . . . . . . . . . . . . 3 63 2. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 5 64 2.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . 6 65 2.2. General Issues Regarding Time . . . . . . . . . . . . . . 7 66 3. A Singleton Definition for One-way Delay . . . . . . . . . . 8 67 3.1. Metric Name: . . . . . . . . . . . . . . . . . . . . . . 8 68 3.2. Metric Parameters: . . . . . . . . . . . . . . . . . . . 8 69 3.3. Metric Units: . . . . . . . . . . . . . . . . . . . . . . 8 70 3.4. Definition: . . . . . . . . . . . . . . . . . . . . . . . 8 71 3.5. Discussion: . . . . . . . . . . . . . . . . . . . . . . . 9 72 3.6. Methodologies: . . . . . . . . . . . . . . . . . . . . . 10 73 3.7. Errors and Uncertainties: . . . . . . . . . . . . . . . . 11 74 3.7.1. Errors or uncertainties related to Clocks . . . . . . 11 75 3.7.2. Errors or uncertainties related to Wire-time vs Host- 76 time . . . . . . . . . . . . . . . . . . . . . . . . 12 77 3.7.3. Calibration . . . . . . . . . . . . . . . . . . . . . 13 78 3.8. Reporting the metric: . . . . . . . . . . . . . . . . . . 15 79 3.8.1. Type-P . . . . . . . . . . . . . . . . . . . . . . . 15 80 3.8.2. Loss Threshold . . . . . . . . . . . . . . . . . . . 16 81 3.8.3. Calibration Results . . . . . . . . . . . . . . . . . 16 82 3.8.4. Path . . . . . . . . . . . . . . . . . . . . . . . . 16 83 4. A Definition for Samples of One-way Delay . . . . . . . . . . 16 84 4.1. Metric Name: . . . . . . . . . . . . . . . . . . . . . . 17 85 4.2. Metric Parameters: . . . . . . . . . . . . . . . . . . . 17 86 4.3. Metric Units: . . . . . . . . . . . . . . . . . . . . . . 17 87 4.4. Definition: . . . . . . . . . . . . . . . . . . . . . . . 17 88 4.5. Discussion: . . . . . . . . . . . . . . . . . . . . . . . 18 89 4.6. Methodologies: . . . . . . . . . . . . . . . . . . . . . 18 90 4.7. Errors and Uncertainties: . . . . . . . . . . . . . . . . 19 91 4.8. Reporting the metric: . . . . . . . . . . . . . . . . . . 19 92 5. Some Statistics Definitions for One-way Delay . . . . . . . . 19 93 5.1. Type-P-One-way-Delay-Percentile . . . . . . . . . . . . . 19 94 5.2. Type-P-One-way-Delay-Median . . . . . . . . . . . . . . . 20 95 5.3. Type-P-One-way-Delay-Minimum . . . . . . . . . . . . . . 21 96 5.4. Type-P-One-way-Delay-Inverse-Percentile . . . . . . . . . 21 97 6. Security Considerations . . . . . . . . . . . . . . . . . . . 21 98 7. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 22 99 8. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 22 100 9. References . . . . . . . . . . . . . . . . . . . . . . . . . 22 101 9.1. Normative References . . . . . . . . . . . . . . . . . . 22 102 9.2. Informative References . . . . . . . . . . . . . . . . . 23 103 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 24 105 1. RFC 2679 bis 107 The following text constitutes RFC 2769 bis proposed for advancement 108 on the IETF Standards Track. This section tracks the changes from 109 [RFC2679]. 111 [RFC6808] provides the test plan and results supporting [RFC2679] 112 advancement along the standards track, according to the process in 113 [RFC6576]. The conclusions of [RFC6808] list four minor 114 modifications: 116 1. Section 6.2.3 of [RFC6808] asserts that the assumption of post- 117 processing to enforce a constant waiting time threshold is 118 compliant, and that the text of the RFC should be revised 119 slightly to include this point (see the last list item of section 120 3.6, below). 122 2. Section 6.5 of [RFC6808] indicates that Type-P-One-way-Delay- 123 Inverse-Percentile statistic has been ignored in both 124 implementations, so it is a candidate for removal or deprecation 125 in RFC2679bis (this small discrepancy does not affect candidacy 126 for advancement) (see section 5.4, below). 128 3. The IETF has reached consensus on guidance for reporting metrics 129 in [RFC6703], and this memo should be referenced in RFC2679bis to 130 incorporate recent experience where appropriate (see the last 131 list item of section 3.6, section 3.8, and section 5 below). 133 4. There is currently one erratum with status "Held for document 134 update" for [RFC2679], and it appears this minor revision and 135 additional text should be incorporated in RFC2679bis (see section 136 5.1). 138 A number of updates to the [RFC2679] text have been implemented in 139 the text below, to reference key IPPM RFCs that were approved after 141 [RFC2679], and to address comments on the IPPM mailing list 142 describing current conditions and experience. 144 1. Near the end of section 2.1, update of a network example using 145 ATM and clarification of TCP's affect on queue occupation and 146 importance of one-way delay measurement. 148 2. Explicit inclusion of the maximum waiting time input parameter 149 in section 3.2 and 4.2, reflecting recognition of this parameter 150 in more recent RFCs and ITU-T Recommendation Y.1540. 152 3. Addition of reference to RFC6703 in the discussion of packet 153 life time and application timeouts in section 3.5. 155 4. Addition of reference to the default requirement (that packets 156 be standard-formed) from RFC2330 as a new list item in section 157 3.5. 159 5. GPS-based NTP experience replaces "to be tested" in section 3.5. 161 6. Added parenthetical guidance on minimizing interval between 162 timestamp placement to send time in section 3.6. 164 7. Added text recognizing the impending deployment of transport 165 layer encryption in section 3.6. 167 8. Section 3.7.2 notes that some current systems perform host time 168 stamping on the network interface hardware. 170 9. "instrument" replaced by the defined term "host" in sections 171 3.7.3 and 3.8.3. 173 10. Added reference to RFC 3432 Periodic sampling alongside Poisson 174 sampling in section 4, and also noting that a truncated Poisson 175 distribution may be needed with modern networks as described in 176 the IPPM Framework update, RFC7312. 178 11. Add reference to RFC 4737 Reordering metric in the related 179 discussion of section 4.6, Methodologies. 181 12. Clarifying the conclusions on two related points on harm to 182 measurements (recognition of measurement traffic for unexpected 183 priority treatment and attacker traffic which emulates 184 measurement) in section 6, Security Considerations. 186 Section 5.4.4 of [RFC6390] suggests a common template for performance 187 metrics partially derived from previous IPPM and BMWG RFCs, but also 188 contains some new items. All of the [RFC6390] Normative points are 189 covered, but not quite in the same section names or orientation. 190 Several of the Informative points are covered. Maintaining the 191 familiar outline of IPPM literature has both value and minimizes 192 unnecessary differences between this revised RFC and current/future 193 IPPM RFCs. 195 The publication of RFC 6921 suggested an area where this memo might 196 need updating. Packet transfer on Faster-Than-Light (FTL) networks 197 could result in negative delays and packet reordering, however both 198 are covered as possibilities in the current text and no revisions are 199 deemed necessary (we also note that this is an April 1st RFC). 201 2. Introduction 203 This memo defines a metric for one-way delay of packets across 204 Internet paths. It builds on notions introduced and discussed in the 205 IPPM Framework document, [RFC2330]; the reader is assumed to be 206 familiar with that document. 208 This memo is intended to be parallel in structure to a companion 209 document for Packet Loss ("A One-way Packet Loss Metric for IPPM") 210 [RFC2680]. 212 Although [RFC2119] was written with protocols in mind, the key words 213 are used in this document for similar reasons. They are used to 214 ensure the results of measurements from two different implementations 215 are comparable, and to note instances when an implementation could 216 perturb the network. 218 The structure of the memo is as follows: 220 + A 'singleton' analytic metric, called Type-P-One-way-Delay, will be 221 introduced to measure a single observation of one-way delay. 223 + Using this singleton metric, a 'sample', called Type-P-One-way- 224 Delay-Poisson-Stream, will be introduced to measure a sequence of 225 singleton delays sent at times taken from a Poisson process. 227 + Using this sample, several 'statistics' of the sample will be 228 defined and discussed. This progression from singleton to sample to 229 statistics, with clear separation among them, is important. 231 Whenever a technical term from the IPPM Framework document is first 232 used in this memo, it will be tagged with a trailing asterisk. For 233 example, "term*" indicates that "term" is defined in the Framework. 235 2.1. Motivation 237 One-way delay of a Type-P* packet from a source host* to a 238 destination host is useful for several reasons: 240 + Some applications do not perform well (or at all) if end-to-end 241 delay between hosts is large relative to some threshold value. 243 + Erratic variation in delay makes it difficult (or impossible) to 244 support many real-time applications. 246 + The larger the value of delay, the more difficult it is for 247 transport-layer protocols to sustain high bandwidths. 249 + The minimum value of this metric provides an indication of the 250 delay due only to propagation and transmission delay. 252 + The minimum value of this metric provides an indication of the 253 delay that will likely be experienced when the path* traversed is 254 lightly loaded. 256 + Values of this metric above the minimum provide an indication of 257 the congestion present in the path. 259 The measurement of one-way delay instead of round-trip delay is 260 motivated by the following factors: 262 + In today's Internet, the path from a source to a destination may be 263 different than the path from the destination back to the source 264 ("asymmetric paths"), such that different sequences of routers are 265 used for the forward and reverse paths. Therefore round-trip 266 measurements actually measure the performance of two distinct paths 267 together. Measuring each path independently highlights the 268 performance difference between the two paths which may traverse 269 different Internet service providers, and even radically different 270 types of networks (for example, research versus commodity networks, 271 or networks with asymmetric link capacities, or wireless vs. wireline 272 access). 274 + Even when the two paths are symmetric, they may have radically 275 different performance characteristics due to asymmetric queueing. 277 + Performance of an application may depend mostly on the performance 278 in one direction. For example, a TCP-based communication may 279 experience reduced throughput if congestion occurs in one direction 280 of its communication. Trouble shooting may be simplified if the 281 congested direction of TCP transmission can be identified. 283 + In quality-of-service (QoS) enabled networks, provisioning in one 284 direction may be radically different than provisioning in the reverse 285 direction, and thus the QoS guarantees differ. Measuring the paths 286 independently allows the verification of both guarantees. 288 It is outside the scope of this document to say precisely how delay 289 metrics would be applied to specific problems. 291 2.2. General Issues Regarding Time 293 {Comment: the terminology below differs from that defined by ITU-T 294 documents (e.g., G.810, "Definitions and terminology for 295 synchronization networks" and I.356, "B-ISDN ATM layer cell transfer 296 performance"), but is consistent with the IPPM Framework document. 297 In general, these differences derive from the different backgrounds; 298 the ITU-T documents historically have a telephony origin, while the 299 authors of this document (and the Framework) have a computer systems 300 background. Although the terms defined below have no direct 301 equivalent in the ITU-T definitions, after our definitions we will 302 provide a rough mapping. However, note one potential confusion: our 303 definition of "clock" is the computer operating systems definition 304 denoting a time-of-day clock, while the ITU-T definition of clock 305 denotes a frequency reference.} 307 Whenever a time (i.e., a moment in history) is mentioned here, it is 308 understood to be measured in seconds (and fractions) relative to UTC. 310 As described more fully in the Framework document, there are four 311 distinct, but related notions of clock uncertainty: 313 synchronization* 315 measures the extent to which two clocks agree on what time it is. 316 For example, the clock on one host might be 5.4 msec ahead of the 317 clock on a second host. {Comment: A rough ITU-T equivalent is "time 318 error".} 320 accuracy* 322 measures the extent to which a given clock agrees with UTC. For 323 example, the clock on a host might be 27.1 msec behind UTC. {Comment: 324 A rough ITU-T equivalent is "time error from UTC".} 326 resolution* 328 measures the precision of a given clock. For example, the clock on 329 an old Unix host might tick only once every 10 msec, and thus have a 330 resolution of only 10 msec. {Comment: A very rough ITU-T equivalent 331 is "sampling period".} 333 skew* 335 measures the change of accuracy, or of synchronization, with time. 336 For example, the clock on a given host might gain 1.3 msec per hour 337 and thus be 27.1 msec behind UTC at one time and only 25.8 msec an 338 hour later. In this case, we say that the clock of the given host 339 has a skew of 1.3 msec per hour relative to UTC, which threatens 340 accuracy. We might also speak of the skew of one clock relative to 341 another clock, which threatens synchronization. {Comment: A rough 342 ITU-T equivalent is "time drift".} 344 3. A Singleton Definition for One-way Delay 346 3.1. Metric Name: 348 Type-P-One-way-Delay 350 3.2. Metric Parameters: 352 + Src, the IP address of a host 354 + Dst, the IP address of a host 356 + T, a time 358 + Tmax, a loss threshold waiting time 360 3.3. Metric Units: 362 The value of a Type-P-One-way-Delay is either a real number, or an 363 undefined (informally, infinite) number of seconds. 365 3.4. Definition: 367 For a real number dT, >>the *Type-P-One-way-Delay* from Src to Dst at 368 T is dT<< means that Src sent the first bit of a Type-P packet to Dst 369 at wire-time* T and that Dst received the last bit of that packet at 370 wire-time T+dT. 372 >>The *Type-P-One-way-Delay* from Src to Dst at T is undefined 373 (informally, infinite)<< means that Src sent the first bit of a 374 Type-P packet to Dst at wire-time T and that Dst did not receive that 375 packet (within the loss threshold waiting time, Tmax). 377 Suggestions for what to report along with metric values appear in 378 Section 3.8 after a discussion of the metric, methodologies for 379 measuring the metric, and error analysis. 381 3.5. Discussion: 383 Type-P-One-way-Delay is a relatively simple analytic metric, and one 384 that we believe will afford effective methods of measurement. 386 The following issues are likely to come up in practice: 388 + Real delay values will be positive. Therefore, it does not make 389 sense to report a negative value as a real delay. However, an 390 individual zero or negative delay value might be useful as part of a 391 stream when trying to discover a distribution of a stream of delay 392 values. 394 + Since delay values will often be as low as the 100 usec to 10 msec 395 range, it will be important for Src and Dst to synchronize very 396 closely. GPS systems afford one way to achieve synchronization to 397 within several 10s of usec. Ordinary application of NTP may allow 398 synchronization to within several msec, but this depends on the 399 stability and symmetry of delay properties among those NTP agents 400 used, and this delay is what we are trying to measure. A combination 401 of some GPS-based NTP servers and a conservatively designed and 402 deployed set of other NTP servers should yield good results. This 403 was tested in [RFC6808], where a GPS measurement system's results 404 compared well with a GPS-based NTP synchronized system for the same 405 intercontinental path. 407 + A given methodology will have to include a way to determine whether 408 a delay value is infinite or whether it is merely very large (and the 409 packet is yet to arrive at Dst). As noted by Mahdavi and Paxson 410 [RFC2678], simple upper bounds (such as the 255 seconds theoretical 411 upper bound on the lifetimes of IP packets [RFC0791]) could be used, 412 but good engineering, including an understanding of packet lifetimes, 413 will be needed in practice. {Comment: Note that, for many 414 applications of these metrics, the harm in treating a large delay as 415 infinite might be zero or very small. A TCP data packet, for 416 example, that arrives only after several multiples of the RTT may as 417 well have been lost. See section 4.1.1 of [RFC6703] for examination 418 of unusual packet delays and application performance estimation.} 420 + If the packet is duplicated along the path (or paths) so that 421 multiple non-corrupt copies arrive at the destination, then the 422 packet is counted as received, and the first copy to arrive 423 determines the packet's one-way delay. 425 + If the packet is fragmented and if, for whatever reason, reassembly 426 does not occur, then the packet will be deemed lost. 428 + The packet is standard-formed, the default criteria for all metric 429 definitions defined in Section 15 of [RFC2330], otherwise the packet 430 will be deemed lost. 432 3.6. Methodologies: 434 As with other Type-P-* metrics, the detailed methodology will depend 435 on the Type-P (e.g., protocol number, UDP/TCP port number, size, 436 precedence). 438 Generally, for a given Type-P, the methodology would proceed as 439 follows: 441 + Arrange that Src and Dst are synchronized; that is, that they have 442 clocks that are very closely synchronized with each other and each 443 fairly close to the actual time. 445 + At the Src host, select Src and Dst IP addresses, and form a test 446 packet of Type-P with these addresses. Any 'padding' portion of the 447 packet needed only to make the test packet a given size should be 448 filled with randomized bits to avoid a situation in which the 449 measured delay is lower than it would otherwise be due to compression 450 techniques along the path. Note that use of transport layer 451 encryption will counteract the deployment of network-based analysis 452 and may reduce the adoption of payload optimizations like 453 compression. 455 + At the Dst host, arrange to receive the packet. 457 + At the Src host, place a timestamp in the prepared Type-P packet, 458 and send it towards Dst (ideally minimizing time before sending). 460 + If the packet arrives within a reasonable period of time, take a 461 timestamp as soon as possible upon the receipt of the packet. By 462 subtracting the two timestamps, an estimate of one-way delay can be 463 computed. Error analysis of a given implementation of the method 464 must take into account the closeness of synchronization between Src 465 and Dst. If the delay between Src's timestamp and the actual sending 466 of the packet is known, then the estimate could be adjusted by 467 subtracting this amount; uncertainty in this value must be taken into 468 account in error analysis. Similarly, if the delay between the 469 actual receipt of the packet and Dst's timestamp is known, then the 470 estimate could be adjusted by subtracting this amount; uncertainty in 471 this value must be taken into account in error analysis. See the 472 next section, "Errors and Uncertainties", for a more detailed 473 discussion. 475 + If the packet fails to arrive within a reasonable period of time, 476 Tmax, the one-way delay is taken to be undefined (informally, 477 infinite). Note that the threshold of 'reasonable' is a parameter of 478 the metric. These points are examined in detail in [RFC6703], 479 including analysis preferences to assign undefined delay to packets 480 that fail to arrive with the difficulties emerging from the informal 481 "infinite delay" assignment, and an estimation of an upper bound on 482 waiting time for packets in transit. Further, enforcing a specific 483 constant waiting time on stored singletons of one-way delay is 484 compliant with this specification and may allow the results to serve 485 more than one reporting audience. 487 Issues such as the packet format, the means by which Dst knows when 488 to expect the test packet, and the means by which Src and Dst are 489 synchronized are outside the scope of this document. {Comment: We 490 plan to document elsewhere our own work in describing such more 491 detailed implementation techniques and we encourage others to as 492 well.} 494 3.7. Errors and Uncertainties: 496 The description of any specific measurement method should include an 497 accounting and analysis of various sources of error or uncertainty. 498 The Framework document provides general guidance on this point, but 499 we note here the following specifics related to delay metrics: 501 + Errors or uncertainties due to uncertainties in the clocks of the 502 Src and Dst hosts. 504 + Errors or uncertainties due to the difference between 'wire time' 505 and 'host time'. 507 In addition, the loss threshold may affect the results. Each of 508 these are discussed in more detail below, along with a section 509 ("Calibration") on accounting for these errors and uncertainties. 511 3.7.1. Errors or uncertainties related to Clocks 513 The uncertainty in a measurement of one-way delay is related, in 514 part, to uncertainties in the clocks of the Src and Dst hosts. In 515 the following, we refer to the clock used to measure when the packet 516 was sent from Src as the source clock, we refer to the clock used to 517 measure when the packet was received by Dst as the destination clock, 518 we refer to the observed time when the packet was sent by the source 519 clock as Tsource, and the observed time when the packet was received 520 by the destination clock as Tdest. Alluding to the notions of 521 synchronization, accuracy, resolution, and skew mentioned in the 522 Introduction, we note the following: 524 + Any error in the synchronization between the source clock and the 525 destination clock will contribute to error in the delay measurement. 526 We say that the source clock and the destination clock have a 527 synchronization error of Tsynch if the source clock is Tsynch ahead 528 of the destination clock. Thus, if we know the value of Tsynch 529 exactly, we could correct for clock synchronization by adding Tsynch 530 to the uncorrected value of Tdest-Tsource. 532 + The accuracy of a clock is important only in identifying the time 533 at which a given delay was measured. Accuracy, per se, has no 534 importance to the accuracy of the measurement of delay. When 535 computing delays, we are interested only in the differences between 536 clock values, not the values themselves. 538 + The resolution of a clock adds to uncertainty about any time 539 measured with it. Thus, if the source clock has a resolution of 10 540 msec, then this adds 10 msec of uncertainty to any time value 541 measured with it. We will denote the resolution of the source clock 542 and the destination clock as Rsource and Rdest, respectively. 544 + The skew of a clock is not so much an additional issue as it is a 545 realization of the fact that Tsynch is itself a function of time. 546 Thus, if we attempt to measure or to bound Tsynch, this needs to be 547 done periodically. Over some periods of time, this function can be 548 approximated as a linear function plus some higher order terms; in 549 these cases, one option is to use knowledge of the linear component 550 to correct the clock. Using this correction, the residual Tsynch is 551 made smaller, but remains a source of uncertainty that must be 552 accounted for. We use the function Esynch(t) to denote an upper 553 bound on the uncertainty in synchronization. Thus, |Tsynch(t)| <= 554 Esynch(t). 556 Taking these items together, we note that naive computation Tdest- 557 Tsource will be off by Tsynch(t) +/- (Rsource + Rdest). Using the 558 notion of Esynch(t), we note that these clock-related problems 559 introduce a total uncertainty of Esynch(t)+ Rsource + Rdest. This 560 estimate of total clock-related uncertainty should be included in the 561 error/uncertainty analysis of any measurement implementation. 563 3.7.2. Errors or uncertainties related to Wire-time vs Host-time 565 As we have defined one-way delay, we would like to measure the time 566 between when the test packet leaves the network interface of Src and 567 when it (completely) arrives at the network interface of Dst, and we 568 refer to these as "wire times." If the timings are themselves 569 performed by software on Src and Dst, however, then this software can 570 only directly measure the time between when Src grabs a timestamp 571 just prior to sending the test packet and when Dst grabs a timestamp 572 just after having received the test packet, and we refer to these two 573 points as "host times". 575 We note that some systems perform host time stamping on the network 576 interface hardware, in an attempt to minimize the difference from 577 wire times. 579 To the extent that the difference between wire time and host time is 580 accurately known, this knowledge can be used to correct for host time 581 measurements and the corrected value more accurately estimates the 582 desired (wire time) metric. 584 To the extent, however, that the difference between wire time and 585 host time is uncertain, this uncertainty must be accounted for in an 586 analysis of a given measurement method. We denote by Hsource an 587 upper bound on the uncertainty in the difference between wire time 588 and host time on the Src host, and similarly define Hdest for the Dst 589 host. We then note that these problems introduce a total uncertainty 590 of Hsource+Hdest. This estimate of total wire-vs-host uncertainty 591 should be included in the error/uncertainty analysis of any 592 measurement implementation. 594 3.7.3. Calibration 596 Generally, the measured values can be decomposed as follows: 598 measured value = true value + systematic error + random error 600 If the systematic error (the constant bias in measured values) can be 601 determined, it can be compensated for in the reported results. 603 reported value = measured value - systematic error 605 therefore 607 reported value = true value + random error 609 The goal of calibration is to determine the systematic and random 610 error generated by the hosts themselves in as much detail as 611 possible. At a minimum, a bound ("e") should be found such that the 612 reported value is in the range (true value - e) to (true value + e) 613 at least 95 percent of the time. We call "e" the calibration error 614 for the measurements. It represents the degree to which the values 615 produced by the measurement host are repeatable; that is, how closely 616 an actual delay of 30 ms is reported as 30 ms. {Comment: 95 percent 617 was chosen because (1) some confidence level is desirable to be able 618 to remove outliers, which will be found in measuring any physical 619 property; (2) a particular confidence level should be specified so 620 that the results of independent implementations can be compared; and 621 (3) even with a prototype user-level implementation, 95% was loose 622 enough to exclude outliers.} 624 From the discussion in the previous two sections, the error in 625 measurements could be bounded by determining all the individual 626 uncertainties, and adding them together to form 628 Esynch(t) + Rsource + Rdest + Hsource + Hdest. 630 However, reasonable bounds on both the clock-related uncertainty 631 captured by the first three terms and the host-related uncertainty 632 captured by the last two terms should be possible by careful design 633 techniques and calibrating the hosts using a known, isolated, network 634 in a lab. 636 For example, the clock-related uncertainties are greatly reduced 637 through the use of a GPS time source. The sum of Esynch(t) + Rsource 638 + Rdest is small, and is also bounded for the duration of the 639 measurement because of the global time source. 641 The host-related uncertainties, Hsource + Hdest, could be bounded by 642 connecting two hosts back-to-back with a high-speed serial link or 643 isolated LAN segment. In this case, repeated measurements are 644 measuring the same one-way delay. 646 If the test packets are small, such a network connection has a 647 minimal delay that may be approximated by zero. The measured delay 648 therefore contains only systematic and random error in the 649 measurement hosts. The "average value" of repeated measurements is 650 the systematic error, and the variation is the random error. 652 One way to compute the systematic error, and the random error to a 653 95% confidence is to repeat the experiment many times - at least 654 hundreds of tests. The systematic error would then be the median. 655 The random error could then be found by removing the systematic error 656 from the measured values. The 95% confidence interval would be the 657 range from the 2.5th percentile to the 97.5th percentile of these 658 deviations from the true value. The calibration error "e" could then 659 be taken to be the largest absolute value of these two numbers, plus 660 the clock-related uncertainty. {Comment: as described, this bound is 661 relatively loose since the uncertainties are added, and the absolute 662 value of the largest deviation is used. As long as the resulting 663 value is not a significant fraction of the measured values, it is a 664 reasonable bound. If the resulting value is a significant fraction 665 of the measured values, then more exact methods will be needed to 666 compute the calibration error.} 668 Note that random error is a function of measurement load. For 669 example, if many paths will be measured by one host, this might 670 increase interrupts, process scheduling, and disk I/O (for example, 671 recording the measurements), all of which may increase the random 672 error in measured singletons. Therefore, in addition to minimal load 673 measurements to find the systematic error, calibration measurements 674 should be performed with the same measurement load that the hosts 675 will see in the field. 677 We wish to reiterate that this statistical treatment refers to the 678 calibration of the host; it is used to "calibrate the meter stick" 679 and say how well the meter stick reflects reality. 681 In addition to calibrating the hosts for finite one-way delay, two 682 checks should be made to ensure that packets reported as losses were 683 really lost. First, the threshold for loss should be verified. In 684 particular, ensure the "reasonable" threshold is reasonable: that it 685 is very unlikely a packet will arrive after the threshold value, and 686 therefore the number of packets lost over an interval is not 687 sensitive to the error bound on measurements. Second, consider the 688 possibility that a packet arrives at the network interface, but is 689 lost due to congestion on that interface or to other resource 690 exhaustion (e.g. buffers) in the host. 692 3.8. Reporting the metric: 694 The calibration and context in which the metric is measured MUST be 695 carefully considered, and SHOULD always be reported along with metric 696 results. We now present four items to consider: the Type-P of test 697 packets, the threshold of infinite delay (if any), error calibration, 698 and the path traversed by the test packets. This list is not 699 exhaustive; any additional information that could be useful in 700 interpreting applications of the metrics should also be reported (see 701 [RFC6703] for extensive discussion of reporting considerations for 702 different audiences). 704 3.8.1. Type-P 706 As noted in the Framework document [RFC2330], the value of the metric 707 may depend on the type of IP packets used to make the measurement, or 708 "type-P". The value of Type-P-One-way-Delay could change if the 709 protocol (UDP or TCP), port number, size, or arrangement for special 710 treatment (e.g., IP precedence or RSVP) changes. The exact Type-P 711 used to make the measurements MUST be accurately reported. 713 3.8.2. Loss Threshold 715 In addition, the threshold (or methodology to distinguish) between a 716 large finite delay and loss MUST be reported. 718 3.8.3. Calibration Results 720 + If the systematic error can be determined, it SHOULD be removed 721 from the measured values. 723 + You SHOULD also report the calibration error, e, such that the true 724 value is the reported value plus or minus e, with 95% confidence (see 725 the last section.) 727 + If possible, the conditions under which a test packet with finite 728 delay is reported as lost due to resource exhaustion on the 729 measurement host SHOULD be reported. 731 3.8.4. Path 733 Finally, the path traversed by the packet SHOULD be reported, if 734 possible. In general it is impractical to know the precise path a 735 given packet takes through the network. The precise path may be 736 known for certain Type-P on short or stable paths. If Type-P 737 includes the record route (or loose-source route) option in the IP 738 header, and the path is short enough, and all routers* on the path 739 support record (or loose-source) route, then the path will be 740 precisely recorded. This is impractical because the route must be 741 short enough, many routers do not support (or are not configured for) 742 record route, and use of this feature would often artificially worsen 743 the performance observed by removing the packet from common-case 744 processing. However, partial information is still valuable context. 745 For example, if a host can choose between two links* (and hence two 746 separate routes from Src to Dst), then the initial link used is 747 valuable context. {Comment: For example, with Merit's NetNow setup, a 748 Src on one NAP can reach a Dst on another NAP by either of several 749 different backbone networks.} 751 4. A Definition for Samples of One-way Delay 753 Given the singleton metric Type-P-One-way-Delay, we now define one 754 particular sample of such singletons. The idea of the sample is to 755 select a particular binding of the parameters Src, Dst, and Type-P, 756 then define a sample of values of parameter T. The means for 757 defining the values of T is to select a beginning time T0, a final 758 time Tf, and an average rate lambda, then define a pseudo-random 759 Poisson process of rate lambda, whose values fall between T0 and Tf. 761 The time interval between successive values of T will then average 1/ 762 lambda. 764 Note that Poisson sampling is only one way of defining a sample. 765 Poisson has the advantage of limiting bias, but other methods of 766 sampling will be appropriate for different situations. For example, 767 a truncated Poisson distribution may be needed to avoid reactive 768 network state changes during intervals of inactivity, see section 4.6 769 of [RFC7321]. Sometimes, the goal is sampling with a known bias, and 770 [RFC3432] describes a method for periodic sampling with random start 771 times. 773 4.1. Metric Name: 775 Type-P-One-way-Delay-Poisson-Stream 777 4.2. Metric Parameters: 779 + Src, the IP address of a host 781 + Dst, the IP address of a host 783 + T0, a time 785 + Tf, a time 787 + Tmax, a loss threshold waiting time 789 + lambda, a rate in reciprocal seconds (or parameters for another 790 distribution) 792 4.3. Metric Units: 794 A sequence of pairs; the elements of each pair are: 796 + T, a time, and 798 + dT, either a real number or an undefined number of seconds. 800 The values of T in the sequence are monotonic increasing. Note that 801 T would be a valid parameter to Type-P-One-way-Delay, and that dT 802 would be a valid value of Type-P-One-way-Delay. 804 4.4. Definition: 806 Given T0, Tf, and lambda, we compute a pseudo-random Poisson process 807 beginning at or before T0, with average arrival rate lambda, and 808 ending at or after Tf. Those time values greater than or equal to T0 809 and less than or equal to Tf are then selected. At each of the times 810 in this process, we obtain the value of Type-P-One-way-Delay at this 811 time. The value of the sample is the sequence made up of the 812 resulting pairs. If there are no such pairs, the 813 sequence is of length zero and the sample is said to be empty. 815 4.5. Discussion: 817 The reader should be familiar with the in-depth discussion of Poisson 818 sampling in the Framework document [RFC2330], which includes methods 819 to compute and verify the pseudo-random Poisson process. 821 We specifically do not constrain the value of lambda, except to note 822 the extremes. If the rate is too large, then the measurement traffic 823 will perturb the network, and itself cause congestion. If the rate 824 is too small, then you might not capture interesting network 825 behavior. {Comment: We expect to document our experiences with, and 826 suggestions for, lambda elsewhere, culminating in a "best current 827 practices" document.} 829 Since a pseudo-random number sequence is employed, the sequence of 830 times, and hence the value of the sample, is not fully specified. 831 Pseudo-random number generators of good quality will be needed to 832 achieve the desired qualities. 834 The sample is defined in terms of a Poisson process both to avoid the 835 effects of self-synchronization and also capture a sample that is 836 statistically as unbiased as possible. {Comment: there is, of course, 837 no claim that real Internet traffic arrives according to a Poisson 838 arrival process.} The Poisson process is used to schedule the delay 839 measurements. The test packets will generally not arrive at Dst 840 according to a Poisson distribution, since they are influenced by the 841 network. 843 All the singleton Type-P-One-way-Delay metrics in the sequence will 844 have the same values of Src, Dst, and Type-P. 846 Note also that, given one sample that runs from T0 to Tf, and given 847 new time values T0' and Tf' such that T0 <= T0' <= Tf' <= Tf, the 848 subsequence of the given sample whose time values fall between T0' 849 and Tf' are also a valid Type-P-One-way-Delay-Poisson-Stream sample. 851 4.6. Methodologies: 853 The methodologies follow directly from: 855 + the selection of specific times, using the specified Poisson 856 arrival process, and 857 + the methodologies discussion already given for the singleton Type- 858 P-One-way-Delay metric. 860 Care must, of course, be given to correctly handle out-of-order 861 arrival of test packets; it is possible that the Src could send one 862 test packet at TS[i], then send a second one (later) at TS[i+1], 863 while the Dst could receive the second test packet at TR[i+1], and 864 then receive the first one (later) at TR[i]. Metrics for reordering 865 may be found in [RFC4737]. 867 4.7. Errors and Uncertainties: 869 In addition to sources of errors and uncertainties associated with 870 methods employed to measure the singleton values that make up the 871 sample, care must be given to analyze the accuracy of the Poisson 872 process with respect to the wire-times of the sending of the test 873 packets. Problems with this process could be caused by several 874 things, including problems with the pseudo-random number techniques 875 used to generate the Poisson arrival process, or with jitter in the 876 value of Hsource (mentioned above as uncertainty in the singleton 877 delay metric). The Framework document shows how to use the Anderson- 878 Darling test to verify the accuracy of a Poisson process over small 879 time frames. {Comment: The goal is to ensure that test packets are 880 sent "close enough" to a Poisson schedule, and avoid periodic 881 behavior.} 883 4.8. Reporting the metric: 885 You MUST report the calibration and context for the underlying 886 singletons along with the stream. (See "Reporting the metric" for 887 Type-P-One-way-Delay.) 889 5. Some Statistics Definitions for One-way Delay 891 Given the sample metric Type-P-One-way-Delay-Poisson-Stream, we now 892 offer several statistics of that sample. These statistics are 893 offered mostly to be illustrative of what could be done. See 894 [RFC6703] for additional discussion of statistics that are relevant 895 to different audiences. 897 5.1. Type-P-One-way-Delay-Percentile 899 Given a Type-P-One-way-Delay-Poisson-Stream and a percent X between 900 0% and 100%, the Xth percentile of all the dT values in the Stream. 901 In computing this percentile, undefined values are treated as 902 infinitely large. Note that this means that the percentile could 903 thus be undefined (informally, infinite). In addition, the Type-P- 904 One-way-Delay-Percentile is undefined if the sample is empty. 906 Example: suppose we take a sample and the results are: 908 Stream1 = < 910 912 914 916 918 920 > 922 Then the 50th percentile would be 110 msec, since 90 msec and 100 923 msec are smaller and 500 msec and 'undefined' are larger. See 924 Section 11.3 of [RFC2330] for computing percentiles. 926 Note that if the possibility that a packet with finite delay is 927 reported as lost is significant, then a high percentile (90th or 928 95th) might be reported as infinite instead of finite. 930 5.2. Type-P-One-way-Delay-Median 932 Given a Type-P-One-way-Delay-Poisson-Stream, the median of all the dT 933 values in the Stream. In computing the median, undefined values are 934 treated as infinitely large. As with Type-P-One-way-Delay- 935 Percentile, Type-P-One-way-Delay-Median is undefined if the sample is 936 empty. 938 As noted in the Framework document, the median differs from the 50th 939 percentile only when the sample contains an even number of values, in 940 which case the mean of the two central values is used. 942 Example: suppose we take a sample and the results are: 944 Stream2 = < > 947 Then the median would be 105 msec, the mean of 100 msec and 110 msec, 948 the two central values. 950 5.3. Type-P-One-way-Delay-Minimum 952 Given a Type-P-One-way-Delay-Poisson-Stream, the minimum of all the 953 dT values in the Stream. In computing this, undefined values are 954 treated as infinitely large. Note that this means that the minimum 955 could thus be undefined (informally, infinite) if all the dT values 956 are undefined. In addition, the Type-P-One-way-Delay-Minimum is 957 undefined if the sample is empty. 959 In the above example, the minimum would be 90 msec. 961 5.4. Type-P-One-way-Delay-Inverse-Percentile 963 Note: This statistic is deprecated in this version of the memo 964 because of lack of use. 966 Given a Type-P-One-way-Delay-Poisson-Stream and a time duration 967 threshold, the fraction of all the dT values in the Stream less than 968 or equal to the threshold. The result could be as low as 0% (if all 969 the dT values exceed threshold) or as high as 100%. Type-P-One-way- 970 Delay-Inverse-Percentile is undefined if the sample is empty. 972 In the above example, the Inverse-Percentile of 103 msec would be 973 50%. 975 6. Security Considerations 977 Conducting Internet measurements raises both security and privacy 978 concerns. This memo does not specify an implementation of the 979 metrics, so it does not directly affect the security of the Internet 980 nor of applications which run on the Internet. However, 981 implementations of these metrics must be mindful of security and 982 privacy concerns. 984 There are two types of security concerns: potential harm caused by 985 the measurements, and potential harm to the measurements. The 986 measurements could cause harm because they are active, and inject 987 packets into the network. The measurement parameters MUST be 988 carefully selected so that the measurements inject trivial amounts of 989 additional traffic into the networks they measure. If they inject 990 "too much" traffic, they can skew the results of the measurement, and 991 in extreme cases cause congestion and denial of service. 993 The measurements themselves could be harmed by routers giving 994 measurement traffic a different priority than "normal" traffic, or by 995 an attacker injecting artificial measurement traffic. If routers can 996 recognize measurement traffic and treat it separately, the 997 measurements will not reflect actual user traffic. Therefore, the 998 measurement methodologies SHOULD include appropriate techniques to 999 reduce the probability measurement traffic can be distinguished from 1000 "normal" traffic. 1002 If an attacker injects packets emulating traffic that are accepted as 1003 legitimate, the loss ratio or other measured values could be 1004 corrupted. Authentication techniques, such as digital signatures, 1005 may be used where appropriate to guard against injected traffic 1006 attacks. 1008 The privacy concerns of network measurement are limited by the active 1009 measurements described in this memo. Unlike passive measurements, 1010 there can be no release of existing user data. 1012 7. IANA Considerations 1014 This memo makes no requests of IANA. 1016 8. Acknowledgements 1018 Special thanks are due to Vern Paxson of Lawrence Berkeley Labs for 1019 his helpful comments on issues of clock uncertainty and statistics. 1020 Thanks also to Garry Couch, Will Leland, Andy Scherrer, Sean Shapira, 1021 and Roland Wittig for several useful suggestions. 1023 9. References 1025 9.1. Normative References 1027 [RFC0791] Postel, J., "Internet Protocol", STD 5, RFC 791, September 1028 1981. 1030 [RFC2026] Bradner, S., "The Internet Standards Process -- Revision 1031 3", BCP 9, RFC 2026, October 1996. 1033 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 1034 Requirement Levels", BCP 14, RFC 2119, March 1997. 1036 [RFC2330] Paxson, V., Almes, G., Mahdavi, J., and M. Mathis, 1037 "Framework for IP Performance Metrics", RFC 2330, May 1038 1998. 1040 [RFC2678] Mahdavi, J. and V. Paxson, "IPPM Metrics for Measuring 1041 Connectivity", RFC 2678, September 1999. 1043 [RFC2679] Almes, G., Kalidindi, S., and M. Zekauskas, "A One-way 1044 Delay Metric for IPPM", RFC 2679, September 1999. 1046 [RFC2680] Almes, G., Kalidindi, S., and M. Zekauskas, "A One-way 1047 Packet Loss Metric for IPPM", RFC 2680, September 1999. 1049 [RFC3432] Raisanen, V., Grotefeld, G., and A. Morton, "Network 1050 performance measurement with periodic streams", RFC 3432, 1051 November 2002. 1053 [RFC4656] Shalunov, S., Teitelbaum, B., Karp, A., Boote, J., and M. 1054 Zekauskas, "A One-way Active Measurement Protocol 1055 (OWAMP)", RFC 4656, September 2006. 1057 [RFC5357] Hedayat, K., Krzanowski, R., Morton, A., Yum, K., and J. 1058 Babiarz, "A Two-Way Active Measurement Protocol (TWAMP)", 1059 RFC 5357, October 2008. 1061 [RFC5657] Dusseault, L. and R. Sparks, "Guidance on Interoperation 1062 and Implementation Reports for Advancement to Draft 1063 Standard", BCP 9, RFC 5657, September 2009. 1065 [RFC5835] Morton, A. and S. Van den Berghe, "Framework for Metric 1066 Composition", RFC 5835, April 2010. 1068 [RFC5905] Mills, D., Martin, J., Burbank, J., and W. Kasch, "Network 1069 Time Protocol Version 4: Protocol and Algorithms 1070 Specification", RFC 5905, June 2010. 1072 [RFC6049] Morton, A. and E. Stephan, "Spatial Composition of 1073 Metrics", RFC 6049, January 2011. 1075 [RFC6576] Geib, R., Morton, A., Fardid, R., and A. Steinmitz, "IP 1076 Performance Metrics (IPPM) Standard Advancement Testing", 1077 BCP 176, RFC 6576, March 2012. 1079 [RFC6703] Morton, A., Ramachandran, G., and G. Maguluri, "Reporting 1080 IP Network Performance Metrics: Different Points of View", 1081 RFC 6703, August 2012. 1083 [RFC7321] McGrew, D. and P. Hoffman, "Cryptographic Algorithm 1084 Implementation Requirements and Usage Guidance for 1085 Encapsulating Security Payload (ESP) and Authentication 1086 Header (AH)", RFC 7321, August 2014. 1088 9.2. Informative References 1090 [ADK] Scholz, F. and M. Stephens, "K-sample Anderson-Darling 1091 Tests of fit, for continuous and discrete cases", 1092 University of Washington, Technical Report No. 81, May 1093 1986. 1095 [I-D.ietf-ippm-testplan-rfc2679] 1096 Ciavattone, L., Geib, R., Morton, A., and M. Wieser, "Test 1097 Plan and Results Supporting Advancement of RFC 2679 on the 1098 Standards Track", draft-ietf-ippm-testplan-rfc2679-03 1099 (work in progress), September 2012. 1101 [RFC3931] Lau, J., Townsley, M., and I. Goyret, "Layer Two Tunneling 1102 Protocol - Version 3 (L2TPv3)", RFC 3931, March 2005. 1104 [RFC4737] Morton, A., Ciavattone, L., Ramachandran, G., Shalunov, 1105 S., and J. Perser, "Packet Reordering Metrics", RFC 4737, 1106 November 2006. 1108 [RFC6390] Clark, A. and B. Claise, "Guidelines for Considering New 1109 Performance Metric Development", BCP 170, RFC 6390, 1110 October 2011. 1112 [RFC6808] Ciavattone, L., Geib, R., Morton, A., and M. Wieser, "Test 1113 Plan and Results Supporting Advancement of RFC 2679 on the 1114 Standards Track", RFC 6808, December 2012. 1116 Authors' Addresses 1118 Guy Almes 1119 Texas A&M 1121 Email: galmes@tamu.edu 1123 Sunil Kalidindi 1124 Ixia 1126 Email: skalidindi@ixiacom.com 1128 Matt Zekauskas 1129 Internet2 1131 Email: matt@internet2.edu 1133 Al Morton (editor) 1134 AT&T Labs 1135 200 Laurel Avenue South 1136 Middletown, NJ 07748 1137 USA 1139 Phone: +1 732 420 1571 1140 Fax: +1 732 368 1192 1141 Email: acmorton@att.com 1142 URI: http://home.comcast.net/~acmacm/