idnits 2.17.1 draft-ietf-ippm-2679-bis-04.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 (August 12, 2015) is 3180 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) ** 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reference: RFC 2680 (Obsoleted by RFC 7680) ** Downref: Normative reference to an Informational RFC: RFC 7312 == Outdated reference: A later version (-02) exists of draft-morton-ippm-2330-stdform-typep-00 Summary: 4 errors (**), 0 flaws (~~), 2 warnings (==), 2 comments (--).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: February 13, 2016 M. Zekauskas 7 Internet2 8 A. Morton, Ed. 9 AT&T Labs 10 August 12, 2015 12 A One-Way Delay Metric for IPPM 13 draft-ietf-ippm-2679-bis-04 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. This memo makes RFC 2679 21 obsolete. 23 Status of This Memo 25 This Internet-Draft is submitted in full conformance with the 26 provisions of BCP 78 and BCP 79. 28 Internet-Drafts are working documents of the Internet Engineering 29 Task Force (IETF). Note that other groups may also distribute 30 working documents as Internet-Drafts. The list of current Internet- 31 Drafts is at http://datatracker.ietf.org/drafts/current/. 33 Internet-Drafts are draft documents valid for a maximum of six months 34 and may be updated, replaced, or obsoleted by other documents at any 35 time. It is inappropriate to use Internet-Drafts as reference 36 material or to cite them other than as "work in progress." 38 This Internet-Draft will expire on February 13, 2016. 40 Copyright Notice 42 Copyright (c) 2015 IETF Trust and the persons identified as the 43 document authors. All rights reserved. 45 This document is subject to BCP 78 and the IETF Trust's Legal 46 Provisions Relating to IETF Documents 47 (http://trustee.ietf.org/license-info) in effect on the date of 48 publication of this document. Please review these documents 49 carefully, as they describe your rights and restrictions with respect 50 to this document. Code Components extracted from this document must 51 include Simplified BSD License text as described in Section 4.e of 52 the Trust Legal Provisions and are provided without warranty as 53 described in the Simplified BSD License. 55 Table of Contents 57 1. Changes from RFC 2679 . . . . . . . . . . . . . . . . . . . . 3 58 2. Introduction . . . . . . . . . . . . . . . . . . . . . . . . 5 59 2.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . 6 60 2.2. General Issues Regarding Time . . . . . . . . . . . . . . 7 61 3. A Singleton Definition for One-way Delay . . . . . . . . . . 8 62 3.1. Metric Name: . . . . . . . . . . . . . . . . . . . . . . 8 63 3.2. Metric Parameters: . . . . . . . . . . . . . . . . . . . 8 64 3.3. Metric Units: . . . . . . . . . . . . . . . . . . . . . . 8 65 3.4. Definition: . . . . . . . . . . . . . . . . . . . . . . . 8 66 3.5. Discussion: . . . . . . . . . . . . . . . . . . . . . . . 9 67 3.6. Methodologies: . . . . . . . . . . . . . . . . . . . . . 10 68 3.7. Errors and Uncertainties: . . . . . . . . . . . . . . . . 11 69 3.7.1. Errors or uncertainties related to Clocks . . . . . . 11 70 3.7.2. Errors or uncertainties related to Wire-time vs Host- 71 time . . . . . . . . . . . . . . . . . . . . . . . . 13 72 3.7.3. Calibration . . . . . . . . . . . . . . . . . . . . . 13 73 3.8. Reporting the metric: . . . . . . . . . . . . . . . . . . 15 74 3.8.1. Type-P . . . . . . . . . . . . . . . . . . . . . . . 16 75 3.8.2. Loss Threshold . . . . . . . . . . . . . . . . . . . 16 76 3.8.3. Calibration Results . . . . . . . . . . . . . . . . . 16 77 3.8.4. Path . . . . . . . . . . . . . . . . . . . . . . . . 16 78 4. A Definition for Samples of One-way Delay . . . . . . . . . . 17 79 4.1. Metric Name: . . . . . . . . . . . . . . . . . . . . . . 17 80 4.2. Metric Parameters: . . . . . . . . . . . . . . . . . . . 17 81 4.3. Metric Units: . . . . . . . . . . . . . . . . . . . . . . 17 82 4.4. Definition: . . . . . . . . . . . . . . . . . . . . . . . 18 83 4.5. Discussion: . . . . . . . . . . . . . . . . . . . . . . . 18 84 4.6. Methodologies: . . . . . . . . . . . . . . . . . . . . . 19 85 4.7. Errors and Uncertainties: . . . . . . . . . . . . . . . . 19 86 4.8. Reporting the metric: . . . . . . . . . . . . . . . . . . 19 87 5. Some Statistics Definitions for One-way Delay . . . . . . . . 19 88 5.1. Type-P-One-way-Delay-Percentile . . . . . . . . . . . . . 20 89 5.2. Type-P-One-way-Delay-Median . . . . . . . . . . . . . . . 20 90 5.3. Type-P-One-way-Delay-Minimum . . . . . . . . . . . . . . 21 91 5.4. Type-P-One-way-Delay-Inverse-Percentile . . . . . . . . . 21 92 6. Security Considerations . . . . . . . . . . . . . . . . . . . 21 93 7. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 22 94 8. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . 22 95 9. References . . . . . . . . . . . . . . . . . . . . . . . . . 23 96 9.1. Normative References . . . . . . . . . . . . . . . . . . 23 97 9.2. Informative References . . . . . . . . . . . . . . . . . 24 98 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 24 100 1. Changes from RFC 2679 102 Note: This section's placement currently preserves minimal 103 differencer between this memo and RFC 2679. The RFC Editor should 104 place this section in an appropriate place. 106 The following text constitutes RFC 2769 bis proposed for advancement 107 on the IETF Standards Track. This section tracks the changes from 108 [RFC2679]. 110 [RFC6808] provides the test plan and results supporting [RFC2679] 111 advancement along the standards track, according to the process in 112 [RFC6576]. The conclusions of [RFC6808] list four minor 113 modifications: 115 1. Section 6.2.3 of [RFC6808] asserts that the assumption of post- 116 processing to enforce a constant waiting time threshold is 117 compliant, and that the text of the RFC should be revised 118 slightly to include this point. The applicability of post- 119 processing was added in the last list item of section 3.6, below. 121 2. Section 6.5 of [RFC6808] indicates that Type-P-One-way-Delay- 122 Inverse-Percentile statistic has been ignored in both 123 implementations, so it is a candidate for removal or deprecation 124 in RFC2679bis (this small discrepancy does not affect candidacy 125 for advancement). This statistic was deprecated in section 5.4, 126 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. This reference 131 was added in the last list item of section 3.6, section 3.8, and 132 in section 5 below. 134 4. There is currently one erratum with status "Held for document 135 update" for [RFC2679], and this minor revision and additional 136 text was incorporated in RFC2679bis in section 5.1, below. 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 140 [RFC2679], and to address comments on the IPPM mailing list 141 describing current conditions and experience. 143 1. Near the end of section 2.1, update of a network example using 144 ATM and clarification of TCP's affect on queue occupation and 145 importance of one-way delay measurement. 147 2. Explicit inclusion of the maximum waiting time input parameter 148 in section 3.2 and 4.2, reflecting recognition of this parameter 149 in more recent RFCs and ITU-T Recommendation Y.1540. 151 3. Addition of reference to RFC6703 in the discussion of packet 152 life time and application timeouts in section 3.5. 154 4. Addition of reference to the default requirement (that packets 155 be standard-formed) from RFC2330 as a new list item in section 156 3.5. 158 5. GPS-based NTP experience replaces "to be tested" in section 3.5. 160 6. Replaced "precedence" with updated terminology (DS Field) in 3.6 161 and 3.8.1 (with reference). 163 7. Added parenthetical guidance on minimizing interval between 164 timestamp placement to send time in section 3.6. 166 8. Added text recognizing the impending deployment of transport 167 layer encryption in section 3.6. 169 9. Section 3.7.2 notes that some current systems perform host time 170 stamping on the network interface hardware. 172 10. "instrument" replaced by the defined term "host" in sections 173 3.7.3 and 3.8.3. 175 11. Added reference to RFC 3432 Periodic sampling alongside Poisson 176 sampling in section 4, and also noting that a truncated Poisson 177 distribution may be needed with modern networks as described in 178 the IPPM Framework update, RFC7312. 180 12. Add reference to RFC 4737 Reordering metric in the related 181 discussion of section 4.6, Methodologies. 183 13. Formatting of Example in section 5.1 modified to match the 184 original (issue with conversion to XML in bis version). 186 14. Clarifying the conclusions on two related points on harm to 187 measurements (recognition of measurement traffic for unexpected 188 priority treatment and attacker traffic which emulates 189 measurement) in section 6, Security Considerations. 191 Section 5.4.4 of [RFC6390] suggests a common template for performance 192 metrics partially derived from previous IPPM and BMWG RFCs, but also 193 contains some new items. All of the [RFC6390] Normative points are 194 covered, but not quite in the same section names or orientation. 195 Several of the Informative points are covered. Maintaining the 196 familiar outline of IPPM literature has both value and minimizes 197 unnecessary differences between this revised RFC and current/future 198 IPPM RFCs. 200 The publication of RFC 6921 suggested an area where this memo might 201 need updating. Packet transfer on Faster-Than-Light (FTL) networks 202 could result in negative delays and packet reordering, however both 203 are covered as possibilities in the current text and no revisions are 204 deemed necessary (we also note that this is an April 1st RFC). 206 2. Introduction 208 This memo defines a metric for one-way delay of packets across 209 Internet paths. It builds on notions introduced and discussed in the 210 IPPM Framework document, [RFC2330]; the reader is assumed to be 211 familiar with that document, and its recent update [RFC7312]. 213 This memo is intended to be parallel in structure to a companion 214 document for Packet Loss ("A One-way Packet Loss Metric for IPPM") 215 [RFC2680]. 217 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 218 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 219 document are to be interpreted as described in RFC 2119 [RFC2119]. 220 Although [RFC2119] was written with protocols in mind, the key words 221 are used in this document for similar reasons. They are used to 222 ensure the results of measurements from two different implementations 223 are comparable, and to note instances when an implementation could 224 perturb the network. 226 The structure of the memo is as follows: 228 + A 'singleton' analytic metric, called Type-P-One-way-Delay, will be 229 introduced to measure a single observation of one-way delay. 231 + Using this singleton metric, a 'sample', called Type-P-One-way- 232 Delay-Poisson-Stream, will be introduced to measure a sequence of 233 singleton delays sent at times taken from a Poisson process. 235 + Using this sample, several 'statistics' of the sample will be 236 defined and discussed. This progression from singleton to sample to 237 statistics, with clear separation among them, is important. 239 Whenever a technical term from the IPPM Framework document is first 240 used in this memo, it will be tagged with a trailing asterisk. For 241 example, "term*" indicates that "term" is defined in the Framework. 243 2.1. Motivation 245 One-way delay of a Type-P* packet from a source host* to a 246 destination host is useful for several reasons: 248 + Some applications do not perform well (or at all) if end-to-end 249 delay between hosts is large relative to some threshold value. 251 + Erratic variation in delay makes it difficult (or impossible) to 252 support many real-time applications. 254 + The larger the value of delay, the more difficult it is for 255 transport-layer protocols to sustain high bandwidths. 257 + The minimum value of this metric provides an indication of the 258 delay due only to propagation and transmission delay. 260 + The minimum value of this metric provides an indication of the 261 delay that will likely be experienced when the path* traversed is 262 lightly loaded. 264 + Values of this metric above the minimum provide an indication of 265 the congestion present in the path. 267 The measurement of one-way delay instead of round-trip delay is 268 motivated by the following factors: 270 + In today's Internet, the path from a source to a destination may be 271 different than the path from the destination back to the source 272 ("asymmetric paths"), such that different sequences of routers are 273 used for the forward and reverse paths. Therefore round-trip 274 measurements actually measure the performance of two distinct paths 275 together. Measuring each path independently highlights the 276 performance difference between the two paths which may traverse 277 different Internet service providers, and even radically different 278 types of networks (for example, research versus commodity networks, 279 or networks with asymmetric link capacities, or wireless vs. wireline 280 access). 282 + Even when the two paths are symmetric, they may have radically 283 different performance characteristics due to asymmetric queueing. 285 + Performance of an application may depend mostly on the performance 286 in one direction. For example, a TCP-based communication will 287 experience reduced throughput if congestion occurs in one direction 288 of its communication. Trouble shooting may be simplified if the 289 congested direction of TCP transmission can be identified. 291 + In quality-of-service (QoS) enabled networks, provisioning in one 292 direction may be radically different than provisioning in the reverse 293 direction, and thus the QoS guarantees differ. Measuring the paths 294 independently allows the verification of both guarantees. 296 It is outside the scope of this document to say precisely how delay 297 metrics would be applied to specific problems. 299 2.2. General Issues Regarding Time 301 {Comment: the terminology below differs from that defined by ITU-T 302 documents (e.g., G.810, "Definitions and terminology for 303 synchronization networks" and I.356, "B-ISDN ATM layer cell transfer 304 performance"), but is consistent with the IPPM Framework document. 305 In general, these differences derive from the different backgrounds; 306 the ITU-T documents historically have a telephony origin, while the 307 authors of this document (and the Framework) have a computer systems 308 background. Although the terms defined below have no direct 309 equivalent in the ITU-T definitions, after our definitions we will 310 provide a rough mapping. However, note one potential confusion: our 311 definition of "clock" is the computer operating systems definition 312 denoting a time-of-day clock, while the ITU-T definition of clock 313 denotes a frequency reference.} 315 Whenever a time (i.e., a moment in history) is mentioned here, it is 316 understood to be measured in seconds (and fractions) relative to UTC. 318 As described more fully in the Framework document, there are four 319 distinct, but related notions of clock uncertainty: 321 synchronization* 323 measures the extent to which two clocks agree on what time it is. 324 For example, the clock on one host might be 5.4 msec ahead of the 325 clock on a second host. {Comment: A rough ITU-T equivalent is "time 326 error".} 328 accuracy* 330 measures the extent to which a given clock agrees with UTC. For 331 example, the clock on a host might be 27.1 msec behind UTC. {Comment: 332 A rough ITU-T equivalent is "time error from UTC".} 334 resolution* 335 measures the precision of a given clock. For example, the clock on 336 an old Unix host might tick only once every 10 msec, and thus have a 337 resolution of only 10 msec. {Comment: A very rough ITU-T equivalent 338 is "sampling period".} 340 skew* 342 measures the change of accuracy, or of synchronization, with time. 343 For example, the clock on a given host might gain 1.3 msec per hour 344 and thus be 27.1 msec behind UTC at one time and only 25.8 msec an 345 hour later. In this case, we say that the clock of the given host 346 has a skew of 1.3 msec per hour relative to UTC, which threatens 347 accuracy. We might also speak of the skew of one clock relative to 348 another clock, which threatens synchronization. {Comment: A rough 349 ITU-T equivalent is "time drift".} 351 3. A Singleton Definition for One-way Delay 353 3.1. Metric Name: 355 Type-P-One-way-Delay 357 3.2. Metric Parameters: 359 + Src, the IP address of a host 361 + Dst, the IP address of a host 363 + T, a time 365 + Tmax, a loss threshold waiting time 367 3.3. Metric Units: 369 The value of a Type-P-One-way-Delay is either a real number, or an 370 undefined (informally, infinite) number of seconds. 372 3.4. Definition: 374 For a real number dT, >>the *Type-P-One-way-Delay* from Src to Dst at 375 T is dT<< means that Src sent the first bit of a Type-P packet to Dst 376 at wire-time* T and that Dst received the last bit of that packet at 377 wire-time T+dT. 379 >>The *Type-P-One-way-Delay* from Src to Dst at T is undefined 380 (informally, infinite)<< means that Src sent the first bit of a 381 Type-P packet to Dst at wire-time T and that Dst did not receive that 382 packet (within the loss threshold waiting time, Tmax). 384 Suggestions for what to report along with metric values appear in 385 Section 3.8 after a discussion of the metric, methodologies for 386 measuring the metric, and error analysis. 388 3.5. Discussion: 390 Type-P-One-way-Delay is a relatively simple analytic metric, and one 391 that we believe will afford effective methods of measurement. 393 The following issues are likely to come up in practice: 395 + Real delay values will be positive. Therefore, it does not make 396 sense to report a negative value as a real delay. However, an 397 individual zero or negative delay value might be useful as part of a 398 stream when trying to discover a distribution of a stream of delay 399 values. 401 + Since delay values will often be as low as the 100 usec to 10 msec 402 range, it will be important for Src and Dst to synchronize very 403 closely. GPS systems afford one way to achieve synchronization to 404 within several 10s of usec. Ordinary application of NTP may allow 405 synchronization to within several msec, but this depends on the 406 stability and symmetry of delay properties among those NTP agents 407 used, and this delay is what we are trying to measure. A combination 408 of some GPS-based NTP servers and a conservatively designed and 409 deployed set of other NTP servers should yield good results. This 410 was tested in [RFC6808], where a GPS measurement system's results 411 compared well with a GPS-based NTP synchronized system for the same 412 intercontinental path. 414 + A given methodology will have to include a way to determine whether 415 a delay value is infinite or whether it is merely very large (and the 416 packet is yet to arrive at Dst). As noted by Mahdavi and Paxson 417 [RFC2678], simple upper bounds (such as the 255 seconds theoretical 418 upper bound on the lifetimes of IP packets [RFC0791]) could be used, 419 but good engineering, including an understanding of packet lifetimes, 420 will be needed in practice. {Comment: Note that, for many 421 applications of these metrics, the harm in treating a large delay as 422 infinite might be zero or very small. A TCP data packet, for 423 example, that arrives only after several multiples of the RTT may as 424 well have been lost. See section 4.1.1 of [RFC6703] for examination 425 of unusual packet delays and application performance estimation.} 427 + If the packet is duplicated along the path (or paths) so that 428 multiple non-corrupt copies arrive at the destination, then the 429 packet is counted as received, and the first copy to arrive 430 determines the packet's one-way delay. 432 + If the packet is fragmented and if, for whatever reason, reassembly 433 does not occur, then the packet will be deemed lost. 435 + The packet is standard-formed, the default criteria for all metric 436 definitions defined in Section 15 of [RFC2330], otherwise the packet 437 will be deemed lost. Note: At this time, the definition of standard- 438 formed packets only applies to IPv4, but also see 439 [I-D.morton-ippm-2330-stdform-typep]. 441 3.6. Methodologies: 443 As with other Type-P-* metrics, the detailed methodology will depend 444 on the Type-P (e.g., protocol number, UDP/TCP port number, size, 445 Differentiated Services (DS) Field [RFC2780]). 447 Generally, for a given Type-P, the methodology would proceed as 448 follows: 450 + Arrange that Src and Dst are synchronized; that is, that they have 451 clocks that are very closely synchronized with each other and each 452 fairly close to the actual time. 454 + At the Src host, select Src and Dst IP addresses, and form a test 455 packet of Type-P with these addresses. Any 'padding' portion of the 456 packet needed only to make the test packet a given size should be 457 filled with randomized bits to avoid a situation in which the 458 measured delay is lower than it would otherwise be due to compression 459 techniques along the path. Note that use of transport layer 460 encryption will counteract the deployment of network-based analysis 461 and may reduce the adoption of network-based payload optimizations 462 like compression. 464 + At the Dst host, arrange to receive the packet. 466 + At the Src host, place a timestamp in the prepared Type-P packet, 467 and send it towards Dst (ideally minimizing time before sending). 469 + If the packet arrives within a reasonable period of time, take a 470 timestamp as soon as possible upon the receipt of the packet. By 471 subtracting the two timestamps, an estimate of one-way delay can be 472 computed. Error analysis of a given implementation of the method 473 must take into account the closeness of synchronization between Src 474 and Dst. If the delay between Src's timestamp and the actual sending 475 of the packet is known, then the estimate could be adjusted by 476 subtracting this amount; uncertainty in this value must be taken into 477 account in error analysis. Similarly, if the delay between the 478 actual receipt of the packet and Dst's timestamp is known, then the 479 estimate could be adjusted by subtracting this amount; uncertainty in 480 this value must be taken into account in error analysis. See the 481 next section, "Errors and Uncertainties", for a more detailed 482 discussion. 484 + If the packet fails to arrive within a reasonable period of time, 485 Tmax, the one-way delay is taken to be undefined (informally, 486 infinite). Note that the threshold of 'reasonable' is a parameter of 487 the metric. These points are examined in detail in [RFC6703], 488 including analysis preferences to assign undefined delay to packets 489 that fail to arrive with the difficulties emerging from the informal 490 "infinite delay" assignment, and an estimation of an upper bound on 491 waiting time for packets in transit. Further, enforcing a specific 492 constant waiting time on stored singletons of one-way delay is 493 compliant with this specification and may allow the results to serve 494 more than one reporting audience. 496 Issues such as the packet format, the means by which Dst knows when 497 to expect the test packet, and the means by which Src and Dst are 498 synchronized are outside the scope of this document. {Comment: We 499 plan to document elsewhere our own work in describing such more 500 detailed implementation techniques and we encourage others to as 501 well.} 503 3.7. Errors and Uncertainties: 505 The description of any specific measurement method should include an 506 accounting and analysis of various sources of error or uncertainty. 507 The Framework document provides general guidance on this point, but 508 we note here the following specifics related to delay metrics: 510 + Errors or uncertainties due to uncertainties in the clocks of the 511 Src and Dst hosts. 513 + Errors or uncertainties due to the difference between 'wire time' 514 and 'host time'. 516 In addition, the loss threshold may affect the results. Each of 517 these are discussed in more detail below, along with a section 518 ("Calibration") on accounting for these errors and uncertainties. 520 3.7.1. Errors or uncertainties related to Clocks 522 The uncertainty in a measurement of one-way delay is related, in 523 part, to uncertainties in the clocks of the Src and Dst hosts. In 524 the following, we refer to the clock used to measure when the packet 525 was sent from Src as the source clock, we refer to the clock used to 526 measure when the packet was received by Dst as the destination clock, 527 we refer to the observed time when the packet was sent by the source 528 clock as Tsource, and the observed time when the packet was received 529 by the destination clock as Tdest. Alluding to the notions of 530 synchronization, accuracy, resolution, and skew mentioned in the 531 Introduction, we note the following: 533 + Any error in the synchronization between the source clock and the 534 destination clock will contribute to error in the delay measurement. 535 We say that the source clock and the destination clock have a 536 synchronization error of Tsynch if the source clock is Tsynch ahead 537 of the destination clock. Thus, if we know the value of Tsynch 538 exactly, we could correct for clock synchronization by adding Tsynch 539 to the uncorrected value of Tdest-Tsource. 541 + The accuracy of a clock is important only in identifying the time 542 at which a given delay was measured. Accuracy, per se, has no 543 importance to the accuracy of the measurement of delay. When 544 computing delays, we are interested only in the differences between 545 clock values, not the values themselves. 547 + The resolution of a clock adds to uncertainty about any time 548 measured with it. Thus, if the source clock has a resolution of 10 549 msec, then this adds 10 msec of uncertainty to any time value 550 measured with it. We will denote the resolution of the source clock 551 and the destination clock as Rsource and Rdest, respectively. 553 + The skew of a clock is not so much an additional issue as it is a 554 realization of the fact that Tsynch is itself a function of time. 555 Thus, if we attempt to measure or to bound Tsynch, this needs to be 556 done periodically. Over some periods of time, this function can be 557 approximated as a linear function plus some higher order terms; in 558 these cases, one option is to use knowledge of the linear component 559 to correct the clock. Using this correction, the residual Tsynch is 560 made smaller, but remains a source of uncertainty that must be 561 accounted for. We use the function Esynch(t) to denote an upper 562 bound on the uncertainty in synchronization. Thus, |Tsynch(t)| <= 563 Esynch(t). 565 Taking these items together, we note that naive computation Tdest- 566 Tsource will be off by Tsynch(t) +/- (Rsource + Rdest). Using the 567 notion of Esynch(t), we note that these clock-related problems 568 introduce a total uncertainty of Esynch(t)+ Rsource + Rdest. This 569 estimate of total clock-related uncertainty should be included in the 570 error/uncertainty analysis of any measurement implementation. 572 3.7.2. Errors or uncertainties related to Wire-time vs Host-time 574 As we have defined one-way delay, we would like to measure the time 575 between when the test packet leaves the network interface of Src and 576 when it (completely) arrives at the network interface of Dst, and we 577 refer to these as "wire times." If the timings are themselves 578 performed by software on Src and Dst, however, then this software can 579 only directly measure the time between when Src grabs a timestamp 580 just prior to sending the test packet and when Dst grabs a timestamp 581 just after having received the test packet, and we refer to these two 582 points as "host times". 584 We note that some systems perform host time stamping on the network 585 interface hardware, in an attempt to minimize the difference from 586 wire times. 588 To the extent that the difference between wire time and host time is 589 accurately known, this knowledge can be used to correct for host time 590 measurements and the corrected value more accurately estimates the 591 desired (wire time) metric. 593 To the extent, however, that the difference between wire time and 594 host time is uncertain, this uncertainty must be accounted for in an 595 analysis of a given measurement method. We denote by Hsource an 596 upper bound on the uncertainty in the difference between wire time 597 and host time on the Src host, and similarly define Hdest for the Dst 598 host. We then note that these problems introduce a total uncertainty 599 of Hsource+Hdest. This estimate of total wire-vs-host uncertainty 600 should be included in the error/uncertainty analysis of any 601 measurement implementation. 603 3.7.3. Calibration 605 Generally, the measured values can be decomposed as follows: 607 measured value = true value + systematic error + random error 609 If the systematic error (the constant bias in measured values) can be 610 determined, it can be compensated for in the reported results. 612 reported value = measured value - systematic error 614 therefore 616 reported value = true value + random error 618 The goal of calibration is to determine the systematic and random 619 error generated by the hosts themselves in as much detail as 620 possible. At a minimum, a bound ("e") should be found such that the 621 reported value is in the range (true value - e) to (true value + e) 622 at least 95 percent of the time. We call "e" the calibration error 623 for the measurements. It represents the degree to which the values 624 produced by the measurement host are repeatable; that is, how closely 625 an actual delay of 30 ms is reported as 30 ms. {Comment: 95 percent 626 was chosen because (1) some confidence level is desirable to be able 627 to remove outliers, which will be found in measuring any physical 628 property; (2) a particular confidence level should be specified so 629 that the results of independent implementations can be compared; and 630 (3) even with a prototype user-level implementation, 95% was loose 631 enough to exclude outliers.} 633 From the discussion in the previous two sections, the error in 634 measurements could be bounded by determining all the individual 635 uncertainties, and adding them together to form 637 Esynch(t) + Rsource + Rdest + Hsource + Hdest. 639 However, reasonable bounds on both the clock-related uncertainty 640 captured by the first three terms and the host-related uncertainty 641 captured by the last two terms should be possible by careful design 642 techniques and calibrating the hosts using a known, isolated, network 643 in a lab. 645 For example, the clock-related uncertainties are greatly reduced 646 through the use of a GPS time source. The sum of Esynch(t) + Rsource 647 + Rdest is small, and is also bounded for the duration of the 648 measurement because of the global time source. 650 The host-related uncertainties, Hsource + Hdest, could be bounded by 651 connecting two hosts back-to-back with a high-speed serial link or 652 isolated LAN segment. In this case, repeated measurements are 653 measuring the same one-way delay. 655 If the test packets are small, such a network connection has a 656 minimal delay that may be approximated by zero. The measured delay 657 therefore contains only systematic and random error in the 658 measurement hosts. The "average value" of repeated measurements is 659 the systematic error, and the variation is the random error. 661 One way to compute the systematic error, and the random error to a 662 95% confidence is to repeat the experiment many times - at least 663 hundreds of tests. The systematic error would then be the median. 664 The random error could then be found by removing the systematic error 665 from the measured values. The 95% confidence interval would be the 666 range from the 2.5th percentile to the 97.5th percentile of these 667 deviations from the true value. The calibration error "e" could then 668 be taken to be the largest absolute value of these two numbers, plus 669 the clock-related uncertainty. {Comment: as described, this bound is 670 relatively loose since the uncertainties are added, and the absolute 671 value of the largest deviation is used. As long as the resulting 672 value is not a significant fraction of the measured values, it is a 673 reasonable bound. If the resulting value is a significant fraction 674 of the measured values, then more exact methods will be needed to 675 compute the calibration error.} 677 Note that random error is a function of measurement load. For 678 example, if many paths will be measured by one host, this might 679 increase interrupts, process scheduling, and disk I/O (for example, 680 recording the measurements), all of which may increase the random 681 error in measured singletons. Therefore, in addition to minimal load 682 measurements to find the systematic error, calibration measurements 683 should be performed with the same measurement load that the hosts 684 will see in the field. 686 We wish to reiterate that this statistical treatment refers to the 687 calibration of the host; it is used to "calibrate the meter stick" 688 and say how well the meter stick reflects reality. 690 In addition to calibrating the hosts for finite one-way delay, two 691 checks should be made to ensure that packets reported as losses were 692 really lost. First, the threshold for loss should be verified. In 693 particular, ensure the "reasonable" threshold is reasonable: that it 694 is very unlikely a packet will arrive after the threshold value, and 695 therefore the number of packets lost over an interval is not 696 sensitive to the error bound on measurements. Second, consider the 697 possibility that a packet arrives at the network interface, but is 698 lost due to congestion on that interface or to other resource 699 exhaustion (e.g. buffers) in the host. 701 3.8. Reporting the metric: 703 The calibration and context in which the metric is measured MUST be 704 carefully considered, and SHOULD always be reported along with metric 705 results. We now present four items to consider: the Type-P of test 706 packets, the threshold of infinite delay (if any), error calibration, 707 and the path traversed by the test packets. This list is not 708 exhaustive; any additional information that could be useful in 709 interpreting applications of the metrics should also be reported (see 710 [RFC6703] for extensive discussion of reporting considerations for 711 different audiences). 713 3.8.1. Type-P 715 As noted in the Framework document, section 13 of [RFC2330], the 716 value of the metric may depend on the type of IP packets used to make 717 the measurement, or "Type-P". The value of Type-P-One-way-Delay 718 could change if the protocol (UDP or TCP), port number, size, or 719 arrangement for special treatment (e.g., IP DS Field [RFC2780], ECN 720 [RFC3168], or RSVP) changes. Additional packet distinctions included 721 in future extensions of the Type-P definition will apply. The exact 722 Type-P used to make the measurements MUST be accurately reported. 724 3.8.2. Loss Threshold 726 In addition, the threshold (or methodology to distinguish) between a 727 large finite delay and loss MUST be reported. 729 3.8.3. Calibration Results 731 + If the systematic error can be determined, it SHOULD be removed 732 from the measured values. 734 + You SHOULD also report the calibration error, e, such that the true 735 value is the reported value plus or minus e, with 95% confidence (see 736 the last section.) 738 + If possible, the conditions under which a test packet with finite 739 delay is reported as lost due to resource exhaustion on the 740 measurement host SHOULD be reported. 742 3.8.4. Path 744 Finally, the path traversed by the packet SHOULD be reported, if 745 possible. In general it is impractical to know the precise path a 746 given packet takes through the network. The precise path may be 747 known for certain Type-P on short or stable paths. If Type-P 748 includes the record route (or loose-source route) option in the IP 749 header, and the path is short enough, and all routers* on the path 750 support record (or loose-source) route, then the path will be 751 precisely recorded. This is impractical because the route must be 752 short enough, many routers do not support (or are not configured for) 753 record route, and use of this feature would often artificially worsen 754 the performance observed by removing the packet from common-case 755 processing. However, partial information is still valuable context. 756 For example, if a host can choose between two links* (and hence two 757 separate routes from Src to Dst), then the initial link used is 758 valuable context. {Comment: For example, with Merit's NetNow setup, a 759 Src on one NAP can reach a Dst on another NAP by either of several 760 different backbone networks.} 762 4. A Definition for Samples of One-way Delay 764 Given the singleton metric Type-P-One-way-Delay, we now define one 765 particular sample of such singletons. The idea of the sample is to 766 select a particular binding of the parameters Src, Dst, and Type-P, 767 then define a sample of values of parameter T. The means for 768 defining the values of T is to select a beginning time T0, a final 769 time Tf, and an average rate lambda, then define a pseudo-random 770 Poisson process of rate lambda, whose values fall between T0 and Tf. 771 The time interval between successive values of T will then average 1/ 772 lambda. 774 Note that Poisson sampling is only one way of defining a sample. 775 Poisson has the advantage of limiting bias, but other methods of 776 sampling will be appropriate for different situations. For example, 777 a truncated Poisson distribution may be needed to avoid reactive 778 network state changes during intervals of inactivity, see section 4.6 779 of [RFC7312]. Sometimes, the goal is sampling with a known bias, and 780 [RFC3432] describes a method for periodic sampling with random start 781 times. 783 4.1. Metric Name: 785 Type-P-One-way-Delay-Poisson-Stream 787 4.2. Metric Parameters: 789 + Src, the IP address of a host 791 + Dst, the IP address of a host 793 + T0, a time 795 + Tf, a time 797 + Tmax, a loss threshold waiting time 799 + lambda, a rate in reciprocal seconds (or parameters for another 800 distribution) 802 4.3. Metric Units: 804 A sequence of pairs; the elements of each pair are: 806 + T, a time, and 808 + dT, either a real number or an undefined number of seconds. 810 The values of T in the sequence are monotonic increasing. Note that 811 T would be a valid parameter to Type-P-One-way-Delay, and that dT 812 would be a valid value of Type-P-One-way-Delay. 814 4.4. Definition: 816 Given T0, Tf, and lambda, we compute a pseudo-random Poisson process 817 beginning at or before T0, with average arrival rate lambda, and 818 ending at or after Tf. Those time values greater than or equal to T0 819 and less than or equal to Tf are then selected. At each of the times 820 in this process, we obtain the value of Type-P-One-way-Delay at this 821 time. The value of the sample is the sequence made up of the 822 resulting pairs. If there are no such pairs, the 823 sequence is of length zero and the sample is said to be empty. 825 4.5. Discussion: 827 The reader should be familiar with the in-depth discussion of Poisson 828 sampling in the Framework document [RFC2330], which includes methods 829 to compute and verify the pseudo-random Poisson process. 831 We specifically do not constrain the value of lambda, except to note 832 the extremes. If the rate is too large, then the measurement traffic 833 will perturb the network, and itself cause congestion. If the rate 834 is too small, then you might not capture interesting network 835 behavior. {Comment: We expect to document our experiences with, and 836 suggestions for, lambda elsewhere, culminating in a "best current 837 practices" document.} 839 Since a pseudo-random number sequence is employed, the sequence of 840 times, and hence the value of the sample, is not fully specified. 841 Pseudo-random number generators of good quality will be needed to 842 achieve the desired qualities. 844 The sample is defined in terms of a Poisson process both to avoid the 845 effects of self-synchronization and also capture a sample that is 846 statistically as unbiased as possible. {Comment: there is, of course, 847 no claim that real Internet traffic arrives according to a Poisson 848 arrival process.} The Poisson process is used to schedule the delay 849 measurements. The test packets will generally not arrive at Dst 850 according to a Poisson distribution, since they are influenced by the 851 network. 853 All the singleton Type-P-One-way-Delay metrics in the sequence will 854 have the same values of Src, Dst, and Type-P. 856 Note also that, given one sample that runs from T0 to Tf, and given 857 new time values T0' and Tf' such that T0 <= T0' <= Tf' <= Tf, the 858 subsequence of the given sample whose time values fall between T0' 859 and Tf' are also a valid Type-P-One-way-Delay-Poisson-Stream sample. 861 4.6. Methodologies: 863 The methodologies follow directly from: 865 + the selection of specific times, using the specified Poisson 866 arrival process, and 868 + the methodologies discussion already given for the singleton Type- 869 P-One-way-Delay metric. 871 Care must, of course, be given to correctly handle out-of-order 872 arrival of test packets; it is possible that the Src could send one 873 test packet at TS[i], then send a second one (later) at TS[i+1], 874 while the Dst could receive the second test packet at TR[i+1], and 875 then receive the first one (later) at TR[i]. Metrics for reordering 876 may be found in [RFC4737]. 878 4.7. Errors and Uncertainties: 880 In addition to sources of errors and uncertainties associated with 881 methods employed to measure the singleton values that make up the 882 sample, care must be given to analyze the accuracy of the Poisson 883 process with respect to the wire-times of the sending of the test 884 packets. Problems with this process could be caused by several 885 things, including problems with the pseudo-random number techniques 886 used to generate the Poisson arrival process, or with jitter in the 887 value of Hsource (mentioned above as uncertainty in the singleton 888 delay metric). The Framework document shows how to use the Anderson- 889 Darling test to verify the accuracy of a Poisson process over small 890 time frames. {Comment: The goal is to ensure that test packets are 891 sent "close enough" to a Poisson schedule, and avoid periodic 892 behavior.} 894 4.8. Reporting the metric: 896 You MUST report the calibration and context for the underlying 897 singletons along with the stream. (See "Reporting the metric" for 898 Type-P-One-way-Delay.) 900 5. Some Statistics Definitions for One-way Delay 902 Given the sample metric Type-P-One-way-Delay-Poisson-Stream, we now 903 offer several statistics of that sample. These statistics are 904 offered mostly to illustrate what could be done. See [RFC6703] for 905 additional discussion of statistics that are relevant to different 906 audiences. 908 5.1. Type-P-One-way-Delay-Percentile 910 Given a Type-P-One-way-Delay-Poisson-Stream and a percent X between 911 0% and 100%, the Xth percentile of all the dT values in the Stream. 912 In computing this percentile, undefined values are treated as 913 infinitely large. Note that this means that the percentile could 914 thus be undefined (informally, infinite). In addition, the Type-P- 915 One-way-Delay-Percentile is undefined if the sample is empty. 917 Example: suppose we take a sample and the results are: 919 Stream1 = < 921 923 925 927 929 931 > 933 Then the 50th percentile would be 110 msec, since 90 msec and 100 934 msec are smaller and 500 msec and 'undefined' are larger. See 935 Section 11.3 of [RFC2330] for computing percentiles. 937 Note that if the possibility that a packet with finite delay is 938 reported as lost is significant, then a high percentile (90th or 939 95th) might be reported as infinite instead of finite. 941 5.2. Type-P-One-way-Delay-Median 943 Given a Type-P-One-way-Delay-Poisson-Stream, the median of all the dT 944 values in the Stream. In computing the median, undefined values are 945 treated as infinitely large. As with Type-P-One-way-Delay- 946 Percentile, Type-P-One-way-Delay-Median is undefined if the sample is 947 empty. 949 As noted in the Framework document, the median differs from the 50th 950 percentile only when the sample contains an even number of values, in 951 which case the mean of the two central values is used. 953 Example: suppose we take a sample and the results are: 955 Stream2 = < 957 959 961 963 965 > 967 Then the median would be 105 msec, the mean of 100 msec and 110 msec, 968 the two central values. 970 5.3. Type-P-One-way-Delay-Minimum 972 Given a Type-P-One-way-Delay-Poisson-Stream, the minimum of all the 973 dT values in the Stream. In computing this, undefined values are 974 treated as infinitely large. Note that this means that the minimum 975 could thus be undefined (informally, infinite) if all the dT values 976 are undefined. In addition, the Type-P-One-way-Delay-Minimum is 977 undefined if the sample is empty. 979 In the above example, the minimum would be 90 msec. 981 5.4. Type-P-One-way-Delay-Inverse-Percentile 983 Note: This statistic is deprecated in this version of the memo 984 because of lack of use. 986 Given a Type-P-One-way-Delay-Poisson-Stream and a time duration 987 threshold, the fraction of all the dT values in the Stream less than 988 or equal to the threshold. The result could be as low as 0% (if all 989 the dT values exceed threshold) or as high as 100%. Type-P-One-way- 990 Delay-Inverse-Percentile is undefined if the sample is empty. 992 In the above example, the Inverse-Percentile of 103 msec would be 993 50%. 995 6. Security Considerations 997 Conducting Internet measurements raises both security and privacy 998 concerns. This memo does not specify an implementation of the 999 metrics, so it does not directly affect the security of the Internet 1000 nor of applications which run on the Internet. However, 1001 implementations of these metrics must be mindful of security and 1002 privacy concerns. 1004 There are two types of security concerns: potential harm caused by 1005 the measurements, and potential harm to the measurements. The 1006 measurements could cause harm because they are active, and inject 1007 packets into the network. The measurement parameters MUST be 1008 carefully selected so that the measurements inject trivial amounts of 1009 additional traffic into the networks they measure. If they inject 1010 "too much" traffic, they can skew the results of the measurement, and 1011 in extreme cases cause congestion and denial of service. 1013 The measurements themselves could be harmed by routers giving 1014 measurement traffic a different priority than "normal" traffic, or by 1015 an attacker injecting artificial measurement traffic. If routers can 1016 recognize measurement traffic and treat it separately, the 1017 measurements will not reflect actual user traffic. Therefore, the 1018 measurement methodologies SHOULD include appropriate techniques to 1019 reduce the probability measurement traffic can be distinguished from 1020 "normal" traffic. 1022 If an attacker injects packets emulating traffic that are accepted as 1023 legitimate, the loss ratio or other measured values could be 1024 corrupted. Authentication techniques, such as digital signatures, 1025 may be used where appropriate to guard against injected traffic 1026 attacks. 1028 The privacy concerns of network measurement are limited by the active 1029 measurements described in this memo. Unlike passive measurements, 1030 there can be no release of existing user data. 1032 7. IANA Considerations 1034 This memo makes no requests of IANA. 1036 8. Acknowledgements 1038 For [RFC2679], special thanks are due to Vern Paxson of Lawrence 1039 Berkeley Labs for his helpful comments on issues of clock uncertainty 1040 and statistics. Thanks also to Garry Couch, Will Leland, Andy 1041 Scherrer, Sean Shapira, and Roland Wittig for several useful 1042 suggestions. 1044 For RFC 2679 bis, thanks to Joachim Fabini, Ruediger Geib, Nalini 1045 Elkins, and Barry Constantine for sharing their measurement 1046 experience as part of their careful reviews. Brian Carpenter and 1047 Scott Bradner provided useful feedback at IETF Last Call. 1049 9. References 1051 9.1. Normative References 1053 [RFC0791] Postel, J., "Internet Protocol", STD 5, RFC 791, 1054 DOI 10.17487/RFC0791, September 1981, 1055 . 1057 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 1058 Requirement Levels", BCP 14, RFC 2119, 1059 DOI 10.17487/RFC2119, March 1997, 1060 . 1062 [RFC2330] Paxson, V., Almes, G., Mahdavi, J., and M. Mathis, 1063 "Framework for IP Performance Metrics", RFC 2330, 1064 DOI 10.17487/RFC2330, May 1998, 1065 . 1067 [RFC2678] Mahdavi, J. and V. Paxson, "IPPM Metrics for Measuring 1068 Connectivity", RFC 2678, DOI 10.17487/RFC2678, September 1069 1999, . 1071 [RFC2679] Almes, G., Kalidindi, S., and M. Zekauskas, "A One-way 1072 Delay Metric for IPPM", RFC 2679, DOI 10.17487/RFC2679, 1073 September 1999, . 1075 [RFC2680] Almes, G., Kalidindi, S., and M. Zekauskas, "A One-way 1076 Packet Loss Metric for IPPM", RFC 2680, 1077 DOI 10.17487/RFC2680, September 1999, 1078 . 1080 [RFC2780] Bradner, S. and V. Paxson, "IANA Allocation Guidelines For 1081 Values In the Internet Protocol and Related Headers", 1082 BCP 37, RFC 2780, DOI 10.17487/RFC2780, March 2000, 1083 . 1085 [RFC3432] Raisanen, V., Grotefeld, G., and A. Morton, "Network 1086 performance measurement with periodic streams", RFC 3432, 1087 DOI 10.17487/RFC3432, November 2002, 1088 . 1090 [RFC6576] Geib, R., Ed., Morton, A., Fardid, R., and A. Steinmitz, 1091 "IP Performance Metrics (IPPM) Standard Advancement 1092 Testing", BCP 176, RFC 6576, DOI 10.17487/RFC6576, March 1093 2012, . 1095 [RFC7312] Fabini, J. and A. Morton, "Advanced Stream and Sampling 1096 Framework for IP Performance Metrics (IPPM)", RFC 7312, 1097 DOI 10.17487/RFC7312, August 2014, 1098 . 1100 9.2. Informative References 1102 [I-D.morton-ippm-2330-stdform-typep] 1103 Morton, A., Fabini, J., Elkins, N., Ackermann, M., and V. 1104 Hegde, "Updates for IPPM's Active Metric Framework: 1105 Packets of Type-P and Standard-Formed Packets", draft- 1106 morton-ippm-2330-stdform-typep-00 (work in progress), 1107 August 2015. 1109 [RFC3168] Ramakrishnan, K., Floyd, S., and D. Black, "The Addition 1110 of Explicit Congestion Notification (ECN) to IP", 1111 RFC 3168, DOI 10.17487/RFC3168, September 2001, 1112 . 1114 [RFC4737] Morton, A., Ciavattone, L., Ramachandran, G., Shalunov, 1115 S., and J. Perser, "Packet Reordering Metrics", RFC 4737, 1116 DOI 10.17487/RFC4737, November 2006, 1117 . 1119 [RFC6390] Clark, A. and B. Claise, "Guidelines for Considering New 1120 Performance Metric Development", BCP 170, RFC 6390, 1121 DOI 10.17487/RFC6390, October 2011, 1122 . 1124 [RFC6703] Morton, A., Ramachandran, G., and G. Maguluri, "Reporting 1125 IP Network Performance Metrics: Different Points of View", 1126 RFC 6703, DOI 10.17487/RFC6703, August 2012, 1127 . 1129 [RFC6808] Ciavattone, L., Geib, R., Morton, A., and M. Wieser, "Test 1130 Plan and Results Supporting Advancement of RFC 2679 on the 1131 Standards Track", RFC 6808, DOI 10.17487/RFC6808, December 1132 2012, . 1134 Authors' Addresses 1136 Guy Almes 1137 Texas A&M 1139 Email: almes@acm.org 1140 Sunil Kalidindi 1141 Ixia 1143 Email: skalidindi@ixiacom.com 1145 Matt Zekauskas 1146 Internet2 1148 Email: matt@internet2.edu 1150 Al Morton (editor) 1151 AT&T Labs 1152 200 Laurel Avenue South 1153 Middletown, NJ 07748 1154 USA 1156 Phone: +1 732 420 1571 1157 Fax: +1 732 368 1192 1158 Email: acmorton@att.com 1159 URI: http://home.comcast.net/~acmacm/