idnits 2.17.1 draft-ietf-ippm-2679-bis-02.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 (June 20, 2015) is 3233 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 Summary: 4 errors (**), 0 flaws (~~), 1 warning (==), 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: December 22, 2015 M. Zekauskas 7 Internet2 8 A. Morton, Ed. 9 AT&T Labs 10 June 20, 2015 12 A One-Way Delay Metric for IPPM 13 draft-ietf-ippm-2679-bis-02 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 December 22, 2015. 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. RFC 2679 bis . . . . . . . . . . . . . . . . . . . . . . . . 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 . . . . . . . . . . . . . . . . . . . . . . . . 12 72 3.7.3. Calibration . . . . . . . . . . . . . . . . . . . . . 13 73 3.8. Reporting the metric: . . . . . . . . . . . . . . . . . . 15 74 3.8.1. Type-P . . . . . . . . . . . . . . . . . . . . . . . 15 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 . . . . . . . . . . 16 79 4.1. Metric Name: . . . . . . . . . . . . . . . . . . . . . . 17 80 4.2. Metric Parameters: . . . . . . . . . . . . . . . . . . . 17 81 4.3. Metric Units: . . . . . . . . . . . . . . . . . . . . . . 17 82 4.4. Definition: . . . . . . . . . . . . . . . . . . . . . . . 17 83 4.5. Discussion: . . . . . . . . . . . . . . . . . . . . . . . 18 84 4.6. Methodologies: . . . . . . . . . . . . . . . . . . . . . 18 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 . . . . . . . . . . . . . 19 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 . . . . . . . . . . . . . . . . . . . . . . . . . 22 96 9.1. Normative References . . . . . . . . . . . . . . . . . . 22 97 9.2. Informative References . . . . . . . . . . . . . . . . . 23 98 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . 23 100 1. RFC 2679 bis 102 The following text constitutes RFC 2769 bis proposed for advancement 103 on the IETF Standards Track. This section tracks the changes from 104 [RFC2679]. 106 [RFC6808] provides the test plan and results supporting [RFC2679] 107 advancement along the standards track, according to the process in 108 [RFC6576]. The conclusions of [RFC6808] list four minor 109 modifications: 111 1. Section 6.2.3 of [RFC6808] asserts that the assumption of post- 112 processing to enforce a constant waiting time threshold is 113 compliant, and that the text of the RFC should be revised 114 slightly to include this point (see the last list item of section 115 3.6, below). 117 2. Section 6.5 of [RFC6808] indicates that Type-P-One-way-Delay- 118 Inverse-Percentile statistic has been ignored in both 119 implementations, so it is a candidate for removal or deprecation 120 in RFC2679bis (this small discrepancy does not affect candidacy 121 for advancement) (see section 5.4, below). 123 3. The IETF has reached consensus on guidance for reporting metrics 124 in [RFC6703], and this memo should be referenced in RFC2679bis to 125 incorporate recent experience where appropriate (see the last 126 list item of section 3.6, section 3.8, and section 5 below). 128 4. There is currently one erratum with status "Held for document 129 update" for [RFC2679], and it appears this minor revision and 130 additional text should be incorporated in RFC2679bis (see section 131 5.1). 133 A number of updates to the [RFC2679] text have been implemented in 134 the text below, to reference key IPPM RFCs that were approved after 135 [RFC2679], and to address comments on the IPPM mailing list 136 describing current conditions and experience. 138 1. Add that RFC2330 was updated by RFC7312 in the Introduction 139 (section 2). 141 2. Near the end of section 2.1, update of a network example using 142 ATM and clarification of TCP's affect on queue occupation and 143 importance of one-way delay measurement. 145 3. Explicit inclusion of the maximum waiting time input parameter 146 in section 3.2 and 4.2, reflecting recognition of this parameter 147 in more recent RFCs and ITU-T Recommendation Y.1540. 149 4. Addition of reference to RFC6703 in the discussion of packet 150 life time and application timeouts in section 3.5. 152 5. Addition of reference to the default requirement (that packets 153 be standard-formed) from RFC2330 as a new list item in section 154 3.5. 156 6. GPS-based NTP experience replaces "to be tested" in section 3.5. 158 7. Replaced "precedence" with updated terminology (DS Field) in 3.6 159 and 3.8.1 (with reference). 161 8. Added parenthetical guidance on minimizing interval between 162 timestamp placement to send time in section 3.6. 164 9. Added text recognizing the impending deployment of transport 165 layer encryption in section 3.6. 167 10. Section 3.7.2 notes that some current systems perform host time 168 stamping on the network interface hardware. 170 11. "instrument" replaced by the defined term "host" in sections 171 3.7.3 and 3.8.3. 173 12. 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 13. Add reference to RFC 4737 Reordering metric in the related 179 discussion of section 4.6, Methodologies. 181 14. Formatting of Example in section 5.1 modified to match the 182 original (issue with conversion to XML in bis version). 184 15. Clarifying the conclusions on two related points on harm to 185 measurements (recognition of measurement traffic for unexpected 186 priority treatment and attacker traffic which emulates 187 measurement) in section 6, Security Considerations. 189 Section 5.4.4 of [RFC6390] suggests a common template for performance 190 metrics partially derived from previous IPPM and BMWG RFCs, but also 191 contains some new items. All of the [RFC6390] Normative points are 192 covered, but not quite in the same section names or orientation. 194 Several of the Informative points are covered. Maintaining the 195 familiar outline of IPPM literature has both value and minimizes 196 unnecessary differences between this revised RFC and current/future 197 IPPM RFCs. 199 The publication of RFC 6921 suggested an area where this memo might 200 need updating. Packet transfer on Faster-Than-Light (FTL) networks 201 could result in negative delays and packet reordering, however both 202 are covered as possibilities in the current text and no revisions are 203 deemed necessary (we also note that this is an April 1st RFC). 205 2. Introduction 207 This memo defines a metric for one-way delay of packets across 208 Internet paths. It builds on notions introduced and discussed in the 209 IPPM Framework document, [RFC2330]; the reader is assumed to be 210 familiar with that document, and its recent update [RFC7312]. 212 This memo is intended to be parallel in structure to a companion 213 document for Packet Loss ("A One-way Packet Loss Metric for IPPM") 214 [RFC2680]. 216 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 217 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 218 document are to be interpreted as described in RFC 2119 [RFC2119]. 219 Although [RFC2119] was written with protocols in mind, the key words 220 are used in this document for similar reasons. They are used to 221 ensure the results of measurements from two different implementations 222 are comparable, and to note instances when an implementation could 223 perturb the network. 225 The structure of the memo is as follows: 227 + A 'singleton' analytic metric, called Type-P-One-way-Delay, will be 228 introduced to measure a single observation of one-way delay. 230 + Using this singleton metric, a 'sample', called Type-P-One-way- 231 Delay-Poisson-Stream, will be introduced to measure a sequence of 232 singleton delays sent at times taken from a Poisson process. 234 + Using this sample, several 'statistics' of the sample will be 235 defined and discussed. This progression from singleton to sample to 236 statistics, with clear separation among them, is important. 238 Whenever a technical term from the IPPM Framework document is first 239 used in this memo, it will be tagged with a trailing asterisk. For 240 example, "term*" indicates that "term" is defined in the Framework. 242 2.1. Motivation 244 One-way delay of a Type-P* packet from a source host* to a 245 destination host is useful for several reasons: 247 + Some applications do not perform well (or at all) if end-to-end 248 delay between hosts is large relative to some threshold value. 250 + Erratic variation in delay makes it difficult (or impossible) to 251 support many real-time applications. 253 + The larger the value of delay, the more difficult it is for 254 transport-layer protocols to sustain high bandwidths. 256 + The minimum value of this metric provides an indication of the 257 delay due only to propagation and transmission delay. 259 + The minimum value of this metric provides an indication of the 260 delay that will likely be experienced when the path* traversed is 261 lightly loaded. 263 + Values of this metric above the minimum provide an indication of 264 the congestion present in the path. 266 The measurement of one-way delay instead of round-trip delay is 267 motivated by the following factors: 269 + In today's Internet, the path from a source to a destination may be 270 different than the path from the destination back to the source 271 ("asymmetric paths"), such that different sequences of routers are 272 used for the forward and reverse paths. Therefore round-trip 273 measurements actually measure the performance of two distinct paths 274 together. Measuring each path independently highlights the 275 performance difference between the two paths which may traverse 276 different Internet service providers, and even radically different 277 types of networks (for example, research versus commodity networks, 278 or networks with asymmetric link capacities, or wireless vs. wireline 279 access). 281 + Even when the two paths are symmetric, they may have radically 282 different performance characteristics due to asymmetric queueing. 284 + Performance of an application may depend mostly on the performance 285 in one direction. For example, a TCP-based communication may 286 experience reduced throughput if congestion occurs in one direction 287 of its communication. Trouble shooting may be simplified if the 288 congested direction of TCP transmission can be identified. 290 + In quality-of-service (QoS) enabled networks, provisioning in one 291 direction may be radically different than provisioning in the reverse 292 direction, and thus the QoS guarantees differ. Measuring the paths 293 independently allows the verification of both guarantees. 295 It is outside the scope of this document to say precisely how delay 296 metrics would be applied to specific problems. 298 2.2. General Issues Regarding Time 300 {Comment: the terminology below differs from that defined by ITU-T 301 documents (e.g., G.810, "Definitions and terminology for 302 synchronization networks" and I.356, "B-ISDN ATM layer cell transfer 303 performance"), but is consistent with the IPPM Framework document. 304 In general, these differences derive from the different backgrounds; 305 the ITU-T documents historically have a telephony origin, while the 306 authors of this document (and the Framework) have a computer systems 307 background. Although the terms defined below have no direct 308 equivalent in the ITU-T definitions, after our definitions we will 309 provide a rough mapping. However, note one potential confusion: our 310 definition of "clock" is the computer operating systems definition 311 denoting a time-of-day clock, while the ITU-T definition of clock 312 denotes a frequency reference.} 314 Whenever a time (i.e., a moment in history) is mentioned here, it is 315 understood to be measured in seconds (and fractions) relative to UTC. 317 As described more fully in the Framework document, there are four 318 distinct, but related notions of clock uncertainty: 320 synchronization* 322 measures the extent to which two clocks agree on what time it is. 323 For example, the clock on one host might be 5.4 msec ahead of the 324 clock on a second host. {Comment: A rough ITU-T equivalent is "time 325 error".} 327 accuracy* 329 measures the extent to which a given clock agrees with UTC. For 330 example, the clock on a host might be 27.1 msec behind UTC. {Comment: 331 A rough ITU-T equivalent is "time error from UTC".} 333 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. 439 3.6. Methodologies: 441 As with other Type-P-* metrics, the detailed methodology will depend 442 on the Type-P (e.g., protocol number, UDP/TCP port number, size, 443 Differentiated Services (DS) Field [RFC2780]). 445 Generally, for a given Type-P, the methodology would proceed as 446 follows: 448 + Arrange that Src and Dst are synchronized; that is, that they have 449 clocks that are very closely synchronized with each other and each 450 fairly close to the actual time. 452 + At the Src host, select Src and Dst IP addresses, and form a test 453 packet of Type-P with these addresses. Any 'padding' portion of the 454 packet needed only to make the test packet a given size should be 455 filled with randomized bits to avoid a situation in which the 456 measured delay is lower than it would otherwise be due to compression 457 techniques along the path. Note that use of transport layer 458 encryption will counteract the deployment of network-based analysis 459 and may reduce the adoption of payload optimizations like 460 compression. 462 + At the Dst host, arrange to receive the packet. 464 + At the Src host, place a timestamp in the prepared Type-P packet, 465 and send it towards Dst (ideally minimizing time before sending). 467 + If the packet arrives within a reasonable period of time, take a 468 timestamp as soon as possible upon the receipt of the packet. By 469 subtracting the two timestamps, an estimate of one-way delay can be 470 computed. Error analysis of a given implementation of the method 471 must take into account the closeness of synchronization between Src 472 and Dst. If the delay between Src's timestamp and the actual sending 473 of the packet is known, then the estimate could be adjusted by 474 subtracting this amount; uncertainty in this value must be taken into 475 account in error analysis. Similarly, if the delay between the 476 actual receipt of the packet and Dst's timestamp is known, then the 477 estimate could be adjusted by subtracting this amount; uncertainty in 478 this value must be taken into account in error analysis. See the 479 next section, "Errors and Uncertainties", for a more detailed 480 discussion. 482 + If the packet fails to arrive within a reasonable period of time, 483 Tmax, the one-way delay is taken to be undefined (informally, 484 infinite). Note that the threshold of 'reasonable' is a parameter of 485 the metric. These points are examined in detail in [RFC6703], 486 including analysis preferences to assign undefined delay to packets 487 that fail to arrive with the difficulties emerging from the informal 488 "infinite delay" assignment, and an estimation of an upper bound on 489 waiting time for packets in transit. Further, enforcing a specific 490 constant waiting time on stored singletons of one-way delay is 491 compliant with this specification and may allow the results to serve 492 more than one reporting audience. 494 Issues such as the packet format, the means by which Dst knows when 495 to expect the test packet, and the means by which Src and Dst are 496 synchronized are outside the scope of this document. {Comment: We 497 plan to document elsewhere our own work in describing such more 498 detailed implementation techniques and we encourage others to as 499 well.} 501 3.7. Errors and Uncertainties: 503 The description of any specific measurement method should include an 504 accounting and analysis of various sources of error or uncertainty. 505 The Framework document provides general guidance on this point, but 506 we note here the following specifics related to delay metrics: 508 + Errors or uncertainties due to uncertainties in the clocks of the 509 Src and Dst hosts. 511 + Errors or uncertainties due to the difference between 'wire time' 512 and 'host time'. 514 In addition, the loss threshold may affect the results. Each of 515 these are discussed in more detail below, along with a section 516 ("Calibration") on accounting for these errors and uncertainties. 518 3.7.1. Errors or uncertainties related to Clocks 520 The uncertainty in a measurement of one-way delay is related, in 521 part, to uncertainties in the clocks of the Src and Dst hosts. In 522 the following, we refer to the clock used to measure when the packet 523 was sent from Src as the source clock, we refer to the clock used to 524 measure when the packet was received by Dst as the destination clock, 525 we refer to the observed time when the packet was sent by the source 526 clock as Tsource, and the observed time when the packet was received 527 by the destination clock as Tdest. Alluding to the notions of 528 synchronization, accuracy, resolution, and skew mentioned in the 529 Introduction, we note the following: 531 + Any error in the synchronization between the source clock and the 532 destination clock will contribute to error in the delay measurement. 533 We say that the source clock and the destination clock have a 534 synchronization error of Tsynch if the source clock is Tsynch ahead 535 of the destination clock. Thus, if we know the value of Tsynch 536 exactly, we could correct for clock synchronization by adding Tsynch 537 to the uncorrected value of Tdest-Tsource. 539 + The accuracy of a clock is important only in identifying the time 540 at which a given delay was measured. Accuracy, per se, has no 541 importance to the accuracy of the measurement of delay. When 542 computing delays, we are interested only in the differences between 543 clock values, not the values themselves. 545 + The resolution of a clock adds to uncertainty about any time 546 measured with it. Thus, if the source clock has a resolution of 10 547 msec, then this adds 10 msec of uncertainty to any time value 548 measured with it. We will denote the resolution of the source clock 549 and the destination clock as Rsource and Rdest, respectively. 551 + The skew of a clock is not so much an additional issue as it is a 552 realization of the fact that Tsynch is itself a function of time. 553 Thus, if we attempt to measure or to bound Tsynch, this needs to be 554 done periodically. Over some periods of time, this function can be 555 approximated as a linear function plus some higher order terms; in 556 these cases, one option is to use knowledge of the linear component 557 to correct the clock. Using this correction, the residual Tsynch is 558 made smaller, but remains a source of uncertainty that must be 559 accounted for. We use the function Esynch(t) to denote an upper 560 bound on the uncertainty in synchronization. Thus, |Tsynch(t)| <= 561 Esynch(t). 563 Taking these items together, we note that naive computation Tdest- 564 Tsource will be off by Tsynch(t) +/- (Rsource + Rdest). Using the 565 notion of Esynch(t), we note that these clock-related problems 566 introduce a total uncertainty of Esynch(t)+ Rsource + Rdest. This 567 estimate of total clock-related uncertainty should be included in the 568 error/uncertainty analysis of any measurement implementation. 570 3.7.2. Errors or uncertainties related to Wire-time vs Host-time 572 As we have defined one-way delay, we would like to measure the time 573 between when the test packet leaves the network interface of Src and 574 when it (completely) arrives at the network interface of Dst, and we 575 refer to these as "wire times." If the timings are themselves 576 performed by software on Src and Dst, however, then this software can 577 only directly measure the time between when Src grabs a timestamp 578 just prior to sending the test packet and when Dst grabs a timestamp 579 just after having received the test packet, and we refer to these two 580 points as "host times". 582 We note that some systems perform host time stamping on the network 583 interface hardware, in an attempt to minimize the difference from 584 wire times. 586 To the extent that the difference between wire time and host time is 587 accurately known, this knowledge can be used to correct for host time 588 measurements and the corrected value more accurately estimates the 589 desired (wire time) metric. 591 To the extent, however, that the difference between wire time and 592 host time is uncertain, this uncertainty must be accounted for in an 593 analysis of a given measurement method. We denote by Hsource an 594 upper bound on the uncertainty in the difference between wire time 595 and host time on the Src host, and similarly define Hdest for the Dst 596 host. We then note that these problems introduce a total uncertainty 597 of Hsource+Hdest. This estimate of total wire-vs-host uncertainty 598 should be included in the error/uncertainty analysis of any 599 measurement implementation. 601 3.7.3. Calibration 603 Generally, the measured values can be decomposed as follows: 605 measured value = true value + systematic error + random error 607 If the systematic error (the constant bias in measured values) can be 608 determined, it can be compensated for in the reported results. 610 reported value = measured value - systematic error 612 therefore 614 reported value = true value + random error 616 The goal of calibration is to determine the systematic and random 617 error generated by the hosts themselves in as much detail as 618 possible. At a minimum, a bound ("e") should be found such that the 619 reported value is in the range (true value - e) to (true value + e) 620 at least 95 percent of the time. We call "e" the calibration error 621 for the measurements. It represents the degree to which the values 622 produced by the measurement host are repeatable; that is, how closely 623 an actual delay of 30 ms is reported as 30 ms. {Comment: 95 percent 624 was chosen because (1) some confidence level is desirable to be able 625 to remove outliers, which will be found in measuring any physical 626 property; (2) a particular confidence level should be specified so 627 that the results of independent implementations can be compared; and 628 (3) even with a prototype user-level implementation, 95% was loose 629 enough to exclude outliers.} 631 From the discussion in the previous two sections, the error in 632 measurements could be bounded by determining all the individual 633 uncertainties, and adding them together to form 635 Esynch(t) + Rsource + Rdest + Hsource + Hdest. 637 However, reasonable bounds on both the clock-related uncertainty 638 captured by the first three terms and the host-related uncertainty 639 captured by the last two terms should be possible by careful design 640 techniques and calibrating the hosts using a known, isolated, network 641 in a lab. 643 For example, the clock-related uncertainties are greatly reduced 644 through the use of a GPS time source. The sum of Esynch(t) + Rsource 645 + Rdest is small, and is also bounded for the duration of the 646 measurement because of the global time source. 648 The host-related uncertainties, Hsource + Hdest, could be bounded by 649 connecting two hosts back-to-back with a high-speed serial link or 650 isolated LAN segment. In this case, repeated measurements are 651 measuring the same one-way delay. 653 If the test packets are small, such a network connection has a 654 minimal delay that may be approximated by zero. The measured delay 655 therefore contains only systematic and random error in the 656 measurement hosts. The "average value" of repeated measurements is 657 the systematic error, and the variation is the random error. 659 One way to compute the systematic error, and the random error to a 660 95% confidence is to repeat the experiment many times - at least 661 hundreds of tests. The systematic error would then be the median. 662 The random error could then be found by removing the systematic error 663 from the measured values. The 95% confidence interval would be the 664 range from the 2.5th percentile to the 97.5th percentile of these 665 deviations from the true value. The calibration error "e" could then 666 be taken to be the largest absolute value of these two numbers, plus 667 the clock-related uncertainty. {Comment: as described, this bound is 668 relatively loose since the uncertainties are added, and the absolute 669 value of the largest deviation is used. As long as the resulting 670 value is not a significant fraction of the measured values, it is a 671 reasonable bound. If the resulting value is a significant fraction 672 of the measured values, then more exact methods will be needed to 673 compute the calibration error.} 675 Note that random error is a function of measurement load. For 676 example, if many paths will be measured by one host, this might 677 increase interrupts, process scheduling, and disk I/O (for example, 678 recording the measurements), all of which may increase the random 679 error in measured singletons. Therefore, in addition to minimal load 680 measurements to find the systematic error, calibration measurements 681 should be performed with the same measurement load that the hosts 682 will see in the field. 684 We wish to reiterate that this statistical treatment refers to the 685 calibration of the host; it is used to "calibrate the meter stick" 686 and say how well the meter stick reflects reality. 688 In addition to calibrating the hosts for finite one-way delay, two 689 checks should be made to ensure that packets reported as losses were 690 really lost. First, the threshold for loss should be verified. In 691 particular, ensure the "reasonable" threshold is reasonable: that it 692 is very unlikely a packet will arrive after the threshold value, and 693 therefore the number of packets lost over an interval is not 694 sensitive to the error bound on measurements. Second, consider the 695 possibility that a packet arrives at the network interface, but is 696 lost due to congestion on that interface or to other resource 697 exhaustion (e.g. buffers) in the host. 699 3.8. Reporting the metric: 701 The calibration and context in which the metric is measured MUST be 702 carefully considered, and SHOULD always be reported along with metric 703 results. We now present four items to consider: the Type-P of test 704 packets, the threshold of infinite delay (if any), error calibration, 705 and the path traversed by the test packets. This list is not 706 exhaustive; any additional information that could be useful in 707 interpreting applications of the metrics should also be reported (see 708 [RFC6703] for extensive discussion of reporting considerations for 709 different audiences). 711 3.8.1. Type-P 713 As noted in the Framework document [RFC2330], the value of the metric 714 may depend on the type of IP packets used to make the measurement, or 715 "type-P". The value of Type-P-One-way-Delay could change if the 716 protocol (UDP or TCP), port number, size, or arrangement for special 717 treatment (e.g., IP DS Field [RFC2780] or RSVP) changes. The exact 718 Type-P used to make the measurements MUST be accurately reported. 720 3.8.2. Loss Threshold 722 In addition, the threshold (or methodology to distinguish) between a 723 large finite delay and loss MUST be reported. 725 3.8.3. Calibration Results 727 + If the systematic error can be determined, it SHOULD be removed 728 from the measured values. 730 + You SHOULD also report the calibration error, e, such that the true 731 value is the reported value plus or minus e, with 95% confidence (see 732 the last section.) 734 + If possible, the conditions under which a test packet with finite 735 delay is reported as lost due to resource exhaustion on the 736 measurement host SHOULD be reported. 738 3.8.4. Path 740 Finally, the path traversed by the packet SHOULD be reported, if 741 possible. In general it is impractical to know the precise path a 742 given packet takes through the network. The precise path may be 743 known for certain Type-P on short or stable paths. If Type-P 744 includes the record route (or loose-source route) option in the IP 745 header, and the path is short enough, and all routers* on the path 746 support record (or loose-source) route, then the path will be 747 precisely recorded. This is impractical because the route must be 748 short enough, many routers do not support (or are not configured for) 749 record route, and use of this feature would often artificially worsen 750 the performance observed by removing the packet from common-case 751 processing. However, partial information is still valuable context. 752 For example, if a host can choose between two links* (and hence two 753 separate routes from Src to Dst), then the initial link used is 754 valuable context. {Comment: For example, with Merit's NetNow setup, a 755 Src on one NAP can reach a Dst on another NAP by either of several 756 different backbone networks.} 758 4. A Definition for Samples of One-way Delay 760 Given the singleton metric Type-P-One-way-Delay, we now define one 761 particular sample of such singletons. The idea of the sample is to 762 select a particular binding of the parameters Src, Dst, and Type-P, 763 then define a sample of values of parameter T. The means for 764 defining the values of T is to select a beginning time T0, a final 765 time Tf, and an average rate lambda, then define a pseudo-random 766 Poisson process of rate lambda, whose values fall between T0 and Tf. 768 The time interval between successive values of T will then average 1/ 769 lambda. 771 Note that Poisson sampling is only one way of defining a sample. 772 Poisson has the advantage of limiting bias, but other methods of 773 sampling will be appropriate for different situations. For example, 774 a truncated Poisson distribution may be needed to avoid reactive 775 network state changes during intervals of inactivity, see section 4.6 776 of [RFC7312]. Sometimes, the goal is sampling with a known bias, and 777 [RFC3432] describes a method for periodic sampling with random start 778 times. 780 4.1. Metric Name: 782 Type-P-One-way-Delay-Poisson-Stream 784 4.2. Metric Parameters: 786 + Src, the IP address of a host 788 + Dst, the IP address of a host 790 + T0, a time 792 + Tf, a time 794 + Tmax, a loss threshold waiting time 796 + lambda, a rate in reciprocal seconds (or parameters for another 797 distribution) 799 4.3. Metric Units: 801 A sequence of pairs; the elements of each pair are: 803 + T, a time, and 805 + dT, either a real number or an undefined number of seconds. 807 The values of T in the sequence are monotonic increasing. Note that 808 T would be a valid parameter to Type-P-One-way-Delay, and that dT 809 would be a valid value of Type-P-One-way-Delay. 811 4.4. Definition: 813 Given T0, Tf, and lambda, we compute a pseudo-random Poisson process 814 beginning at or before T0, with average arrival rate lambda, and 815 ending at or after Tf. Those time values greater than or equal to T0 816 and less than or equal to Tf are then selected. At each of the times 817 in this process, we obtain the value of Type-P-One-way-Delay at this 818 time. The value of the sample is the sequence made up of the 819 resulting pairs. If there are no such pairs, the 820 sequence is of length zero and the sample is said to be empty. 822 4.5. Discussion: 824 The reader should be familiar with the in-depth discussion of Poisson 825 sampling in the Framework document [RFC2330], which includes methods 826 to compute and verify the pseudo-random Poisson process. 828 We specifically do not constrain the value of lambda, except to note 829 the extremes. If the rate is too large, then the measurement traffic 830 will perturb the network, and itself cause congestion. If the rate 831 is too small, then you might not capture interesting network 832 behavior. {Comment: We expect to document our experiences with, and 833 suggestions for, lambda elsewhere, culminating in a "best current 834 practices" document.} 836 Since a pseudo-random number sequence is employed, the sequence of 837 times, and hence the value of the sample, is not fully specified. 838 Pseudo-random number generators of good quality will be needed to 839 achieve the desired qualities. 841 The sample is defined in terms of a Poisson process both to avoid the 842 effects of self-synchronization and also capture a sample that is 843 statistically as unbiased as possible. {Comment: there is, of course, 844 no claim that real Internet traffic arrives according to a Poisson 845 arrival process.} The Poisson process is used to schedule the delay 846 measurements. The test packets will generally not arrive at Dst 847 according to a Poisson distribution, since they are influenced by the 848 network. 850 All the singleton Type-P-One-way-Delay metrics in the sequence will 851 have the same values of Src, Dst, and Type-P. 853 Note also that, given one sample that runs from T0 to Tf, and given 854 new time values T0' and Tf' such that T0 <= T0' <= Tf' <= Tf, the 855 subsequence of the given sample whose time values fall between T0' 856 and Tf' are also a valid Type-P-One-way-Delay-Poisson-Stream sample. 858 4.6. Methodologies: 860 The methodologies follow directly from: 862 + the selection of specific times, using the specified Poisson 863 arrival process, and 864 + the methodologies discussion already given for the singleton Type- 865 P-One-way-Delay metric. 867 Care must, of course, be given to correctly handle out-of-order 868 arrival of test packets; it is possible that the Src could send one 869 test packet at TS[i], then send a second one (later) at TS[i+1], 870 while the Dst could receive the second test packet at TR[i+1], and 871 then receive the first one (later) at TR[i]. Metrics for reordering 872 may be found in [RFC4737]. 874 4.7. Errors and Uncertainties: 876 In addition to sources of errors and uncertainties associated with 877 methods employed to measure the singleton values that make up the 878 sample, care must be given to analyze the accuracy of the Poisson 879 process with respect to the wire-times of the sending of the test 880 packets. Problems with this process could be caused by several 881 things, including problems with the pseudo-random number techniques 882 used to generate the Poisson arrival process, or with jitter in the 883 value of Hsource (mentioned above as uncertainty in the singleton 884 delay metric). The Framework document shows how to use the Anderson- 885 Darling test to verify the accuracy of a Poisson process over small 886 time frames. {Comment: The goal is to ensure that test packets are 887 sent "close enough" to a Poisson schedule, and avoid periodic 888 behavior.} 890 4.8. Reporting the metric: 892 You MUST report the calibration and context for the underlying 893 singletons along with the stream. (See "Reporting the metric" for 894 Type-P-One-way-Delay.) 896 5. Some Statistics Definitions for One-way Delay 898 Given the sample metric Type-P-One-way-Delay-Poisson-Stream, we now 899 offer several statistics of that sample. These statistics are 900 offered mostly to illustrate what could be done. See [RFC6703] for 901 additional discussion of statistics that are relevant to different 902 audiences. 904 5.1. Type-P-One-way-Delay-Percentile 906 Given a Type-P-One-way-Delay-Poisson-Stream and a percent X between 907 0% and 100%, the Xth percentile of all the dT values in the Stream. 908 In computing this percentile, undefined values are treated as 909 infinitely large. Note that this means that the percentile could 910 thus be undefined (informally, infinite). In addition, the Type-P- 911 One-way-Delay-Percentile is undefined if the sample is empty. 913 Example: suppose we take a sample and the results are: 915 Stream1 = < 917 919 921 923 925 927 > 929 Then the 50th percentile would be 110 msec, since 90 msec and 100 930 msec are smaller and 500 msec and 'undefined' are larger. See 931 Section 11.3 of [RFC2330] for computing percentiles. 933 Note that if the possibility that a packet with finite delay is 934 reported as lost is significant, then a high percentile (90th or 935 95th) might be reported as infinite instead of finite. 937 5.2. Type-P-One-way-Delay-Median 939 Given a Type-P-One-way-Delay-Poisson-Stream, the median of all the dT 940 values in the Stream. In computing the median, undefined values are 941 treated as infinitely large. As with Type-P-One-way-Delay- 942 Percentile, Type-P-One-way-Delay-Median is undefined if the sample is 943 empty. 945 As noted in the Framework document, the median differs from the 50th 946 percentile only when the sample contains an even number of values, in 947 which case the mean of the two central values is used. 949 Example: suppose we take a sample and the results are: 951 Stream2 = < 953 955 957 959 960 > 962 Then the median would be 105 msec, the mean of 100 msec and 110 msec, 963 the two central values. 965 5.3. Type-P-One-way-Delay-Minimum 967 Given a Type-P-One-way-Delay-Poisson-Stream, the minimum of all the 968 dT values in the Stream. In computing this, undefined values are 969 treated as infinitely large. Note that this means that the minimum 970 could thus be undefined (informally, infinite) if all the dT values 971 are undefined. In addition, the Type-P-One-way-Delay-Minimum is 972 undefined if the sample is empty. 974 In the above example, the minimum would be 90 msec. 976 5.4. Type-P-One-way-Delay-Inverse-Percentile 978 Note: This statistic is deprecated in this version of the memo 979 because of lack of use. 981 Given a Type-P-One-way-Delay-Poisson-Stream and a time duration 982 threshold, the fraction of all the dT values in the Stream less than 983 or equal to the threshold. The result could be as low as 0% (if all 984 the dT values exceed threshold) or as high as 100%. Type-P-One-way- 985 Delay-Inverse-Percentile is undefined if the sample is empty. 987 In the above example, the Inverse-Percentile of 103 msec would be 988 50%. 990 6. Security Considerations 992 Conducting Internet measurements raises both security and privacy 993 concerns. This memo does not specify an implementation of the 994 metrics, so it does not directly affect the security of the Internet 995 nor of applications which run on the Internet. However, 996 implementations of these metrics must be mindful of security and 997 privacy concerns. 999 There are two types of security concerns: potential harm caused by 1000 the measurements, and potential harm to the measurements. The 1001 measurements could cause harm because they are active, and inject 1002 packets into the network. The measurement parameters MUST be 1003 carefully selected so that the measurements inject trivial amounts of 1004 additional traffic into the networks they measure. If they inject 1005 "too much" traffic, they can skew the results of the measurement, and 1006 in extreme cases cause congestion and denial of service. 1008 The measurements themselves could be harmed by routers giving 1009 measurement traffic a different priority than "normal" traffic, or by 1010 an attacker injecting artificial measurement traffic. If routers can 1011 recognize measurement traffic and treat it separately, the 1012 measurements will not reflect actual user traffic. Therefore, the 1013 measurement methodologies SHOULD include appropriate techniques to 1014 reduce the probability measurement traffic can be distinguished from 1015 "normal" traffic. 1017 If an attacker injects packets emulating traffic that are accepted as 1018 legitimate, the loss ratio or other measured values could be 1019 corrupted. Authentication techniques, such as digital signatures, 1020 may be used where appropriate to guard against injected traffic 1021 attacks. 1023 The privacy concerns of network measurement are limited by the active 1024 measurements described in this memo. Unlike passive measurements, 1025 there can be no release of existing user data. 1027 7. IANA Considerations 1029 This memo makes no requests of IANA. 1031 8. Acknowledgements 1033 For [RFC2679], special thanks are due to Vern Paxson of Lawrence 1034 Berkeley Labs for his helpful comments on issues of clock uncertainty 1035 and statistics. Thanks also to Garry Couch, Will Leland, Andy 1036 Scherrer, Sean Shapira, and Roland Wittig for several useful 1037 suggestions. 1039 For RFC 2679 bis, thanks to Joachim Fabini, Ruediger Geib, Nalini 1040 Elkins, and Barry Constantine for sharing their measurement 1041 experience as part of their careful reviews. 1043 9. References 1045 9.1. Normative References 1047 [RFC0791] Postel, J., "Internet Protocol", STD 5, RFC 791, September 1048 1981. 1050 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 1051 Requirement Levels", BCP 14, RFC 2119, March 1997. 1053 [RFC2330] Paxson, V., Almes, G., Mahdavi, J., and M. Mathis, 1054 "Framework for IP Performance Metrics", RFC 2330, May 1055 1998. 1057 [RFC2678] Mahdavi, J. and V. Paxson, "IPPM Metrics for Measuring 1058 Connectivity", RFC 2678, September 1999. 1060 [RFC2679] Almes, G., Kalidindi, S., and M. Zekauskas, "A One-way 1061 Delay Metric for IPPM", RFC 2679, September 1999. 1063 [RFC2680] Almes, G., Kalidindi, S., and M. Zekauskas, "A One-way 1064 Packet Loss Metric for IPPM", RFC 2680, September 1999. 1066 [RFC2780] Bradner, S. and V. Paxson, "IANA Allocation Guidelines For 1067 Values In the Internet Protocol and Related Headers", BCP 1068 37, RFC 2780, March 2000. 1070 [RFC3432] Raisanen, V., Grotefeld, G., and A. Morton, "Network 1071 performance measurement with periodic streams", RFC 3432, 1072 November 2002. 1074 [RFC6576] Geib, R., Morton, A., Fardid, R., and A. Steinmitz, "IP 1075 Performance Metrics (IPPM) Standard Advancement Testing", 1076 BCP 176, RFC 6576, March 2012. 1078 [RFC7312] Fabini, J. and A. Morton, "Advanced Stream and Sampling 1079 Framework for IP Performance Metrics (IPPM)", RFC 7312, 1080 August 2014. 1082 9.2. Informative References 1084 [RFC4737] Morton, A., Ciavattone, L., Ramachandran, G., Shalunov, 1085 S., and J. Perser, "Packet Reordering Metrics", RFC 4737, 1086 November 2006. 1088 [RFC6390] Clark, A. and B. Claise, "Guidelines for Considering New 1089 Performance Metric Development", BCP 170, RFC 6390, 1090 October 2011. 1092 [RFC6703] Morton, A., Ramachandran, G., and G. Maguluri, "Reporting 1093 IP Network Performance Metrics: Different Points of View", 1094 RFC 6703, August 2012. 1096 [RFC6808] Ciavattone, L., Geib, R., Morton, A., and M. Wieser, "Test 1097 Plan and Results Supporting Advancement of RFC 2679 on the 1098 Standards Track", RFC 6808, December 2012. 1100 Authors' Addresses 1101 Guy Almes 1102 Texas A&M 1104 Email: almes@acm.org 1106 Sunil Kalidindi 1107 Ixia 1109 Email: skalidindi@ixiacom.com 1111 Matt Zekauskas 1112 Internet2 1114 Email: matt@internet2.edu 1116 Al Morton (editor) 1117 AT&T Labs 1118 200 Laurel Avenue South 1119 Middletown, NJ 07748 1120 USA 1122 Phone: +1 732 420 1571 1123 Fax: +1 732 368 1192 1124 Email: acmorton@att.com 1125 URI: http://home.comcast.net/~acmacm/