idnits 2.17.1 draft-morton-ippm-2679-bis-01.txt: Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- No issues found here. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- -- The draft header indicates that this document obsoletes RFC2679, but the abstract doesn't seem to directly say this. It does mention RFC2679 though, so this could be OK. Miscellaneous warnings: ---------------------------------------------------------------------------- == The copyright year in the IETF Trust and authors Copyright Line does not match the current year -- The document date (October 21, 2012) is 4198 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) -- Looks like a reference, but probably isn't: '1' on line 942 -- Looks like a reference, but probably isn't: '2' on line 945 -- Looks like a reference, but probably isn't: '4' on line 950 -- Looks like a reference, but probably isn't: '5' on line 953 -- Looks like a reference, but probably isn't: '3' on line 948 -- Looks like a reference, but probably isn't: '6' on line 955 -- Looks like a reference, but probably isn't: '7' on line 958 == Unused Reference: 'RFC2026' is defined on line 965, but no explicit reference was found in the text == Unused Reference: 'RFC2330' is defined on line 971, but no explicit reference was found in the text == Unused Reference: 'RFC2680' is defined on line 978, but no explicit reference was found in the text == Unused Reference: 'RFC3432' is defined on line 981, but no explicit reference was found in the text == Unused Reference: 'RFC4656' is defined on line 985, but no explicit reference was found in the text == Unused Reference: 'RFC5357' is defined on line 989, but no explicit reference was found in the text == Unused Reference: 'RFC5657' is defined on line 993, but no explicit reference was found in the text == Unused Reference: 'RFC5835' is defined on line 997, but no explicit reference was found in the text == Unused Reference: 'RFC6049' is defined on line 1000, but no explicit reference was found in the text == Unused Reference: 'ADK' is defined on line 1013, but no explicit reference was found in the text == Unused Reference: 'RFC3931' is defined on line 1024, but no explicit reference was found in the text ** Downref: Normative reference to an Informational RFC: RFC 2330 ** Obsolete normative reference: RFC 2679 (Obsoleted by RFC 7679) ** Obsolete normative reference: RFC 2680 (Obsoleted by RFC 7680) ** Downref: Normative reference to an Informational RFC: RFC 5835 ** Downref: Normative reference to an Informational RFC: RFC 6703 Summary: 5 errors (**), 0 flaws (~~), 12 warnings (==), 9 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 Network Working Group G. Almes 3 Internet-Draft Texas A&M 4 Obsoletes: 2679 (if approved) S. Kalidindi 5 Intended status: Standards Track Ixia 6 Expires: April 24, 2013 M. Zekauskas 7 Internet2 8 A. Morton, Ed. 9 AT&T Labs 10 October 21, 2012 12 A One-Way Delay Metric for IPPM 13 draft-morton-ippm-2679-bis-01 15 Abstract 17 This memo (RFC 2679 bis) defines a metric for one-way delay of 18 packets across Internet paths. It builds on notions introduced and 19 discussed in the IPPM Framework document, RFC 2330; the reader is 20 assumed to be familiar with that document. 22 Requirements Language 24 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 25 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 26 document are to be interpreted as described in RFC 2119 [RFC2119]. 28 Status of this Memo 30 This Internet-Draft is submitted in full conformance with the 31 provisions of BCP 78 and BCP 79. 33 Internet-Drafts are working documents of the Internet Engineering 34 Task Force (IETF). Note that other groups may also distribute 35 working documents as Internet-Drafts. The list of current Internet- 36 Drafts is at http://datatracker.ietf.org/drafts/current/. 38 Internet-Drafts are draft documents valid for a maximum of six months 39 and may be updated, replaced, or obsoleted by other documents at any 40 time. It is inappropriate to use Internet-Drafts as reference 41 material or to cite them other than as "work in progress." 43 This Internet-Draft will expire on April 24, 2013. 45 Copyright Notice 47 Copyright (c) 2012 IETF Trust and the persons identified as the 48 document authors. All rights reserved. 50 This document is subject to BCP 78 and the IETF Trust's Legal 51 Provisions Relating to IETF Documents 52 (http://trustee.ietf.org/license-info) in effect on the date of 53 publication of this document. Please review these documents 54 carefully, as they describe your rights and restrictions with respect 55 to this document. Code Components extracted from this document must 56 include Simplified BSD License text as described in Section 4.e of 57 the Trust Legal Provisions and are provided without warranty as 58 described in the Simplified BSD License. 60 Table of Contents 62 1. RFC 2679 bis . . . . . . . . . . . . . . . . . . . . . . . . . 4 63 2. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . 4 64 2.1. Motivation . . . . . . . . . . . . . . . . . . . . . . . . 5 65 2.2. General Issues Regarding Time . . . . . . . . . . . . . . 6 66 3. A Singleton Definition for One-way Delay . . . . . . . . . . . 7 67 3.1. Metric Name: . . . . . . . . . . . . . . . . . . . . . . . 7 68 3.2. Metric Parameters: . . . . . . . . . . . . . . . . . . . . 7 69 3.3. Metric Units: . . . . . . . . . . . . . . . . . . . . . . 8 70 3.4. Definition: . . . . . . . . . . . . . . . . . . . . . . . 8 71 3.5. Discussion: . . . . . . . . . . . . . . . . . . . . . . . 8 72 3.6. Methodologies: . . . . . . . . . . . . . . . . . . . . . . 9 73 3.7. Errors and Uncertainties: . . . . . . . . . . . . . . . . 10 74 3.7.1. Errors or uncertainties related to Clocks . . . . . . 10 75 3.7.2. Errors or uncertainties related to Wire-time vs 76 Host-time . . . . . . . . . . . . . . . . . . . . . . 12 77 3.7.3. Calibration . . . . . . . . . . . . . . . . . . . . . 12 78 3.8. Reporting the metric: . . . . . . . . . . . . . . . . . . 14 79 3.8.1. Type-P . . . . . . . . . . . . . . . . . . . . . . . . 14 80 3.8.2. Loss Threshold . . . . . . . . . . . . . . . . . . . . 15 81 3.8.3. Calibration Results . . . . . . . . . . . . . . . . . 15 82 3.8.4. Path . . . . . . . . . . . . . . . . . . . . . . . . . 15 83 4. A Definition for Samples of One-way Delay . . . . . . . . . . 15 84 4.1. Metric Name: . . . . . . . . . . . . . . . . . . . . . . . 16 85 4.2. Metric Parameters: . . . . . . . . . . . . . . . . . . . . 16 86 4.3. Metric Units: . . . . . . . . . . . . . . . . . . . . . . 16 87 4.4. Definition: . . . . . . . . . . . . . . . . . . . . . . . 16 88 4.5. Discussion: . . . . . . . . . . . . . . . . . . . . . . . 17 89 4.6. Methodologies: . . . . . . . . . . . . . . . . . . . . . . 17 90 4.7. Errors and Uncertainties: . . . . . . . . . . . . . . . . 18 91 4.8. Reporting the metric: . . . . . . . . . . . . . . . . . . 18 92 5. Some Statistics Definitions for One-way Delay . . . . . . . . 18 93 5.1. Type-P-One-way-Delay-Percentile . . . . . . . . . . . . . 18 94 5.2. Type-P-One-way-Delay-Median . . . . . . . . . . . . . . . 19 95 5.3. Type-P-One-way-Delay-Minimum . . . . . . . . . . . . . . . 20 96 5.4. Type-P-One-way-Delay-Inverse-Percentile . . . . . . . . . 20 97 6. Security Considerations . . . . . . . . . . . . . . . . . . . 20 98 7. IANA Considerations . . . . . . . . . . . . . . . . . . . . . 21 99 8. Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . 21 100 9. Refetrences (temporary) . . . . . . . . . . . . . . . . . . . 21 101 10. References . . . . . . . . . . . . . . . . . . . . . . . . . . 22 102 10.1. Normative References . . . . . . . . . . . . . . . . . . . 22 103 10.2. Informative References . . . . . . . . . . . . . . . . . . 23 104 Authors' Addresses . . . . . . . . . . . . . . . . . . . . . . . . 23 106 1. RFC 2679 bis 108 The following text constitutes RFC 2769 bis proposed for advancement 109 on the IETF Standards Track. 111 [I-D.ietf-ippm-testplan-rfc2679] (now approved) provides the test 112 plan and results supporting [RFC2679] advancement along the standards 113 track, according to the process in [RFC6576]. The conclusions of 114 [I-D.ietf-ippm-testplan-rfc2679] list four minor modifications for 115 inclusion: 117 1. Section 6.2.3 of [I-D.ietf-ippm-testplan-rfc2679] asserts that 118 the assumption of post-processing to enforce a constant waiting 119 time threshold is compliant, and that the text of the RFC should 120 be revised slightly to include this point (see the last list item 121 of section 3.6, below). 123 2. Section 6.5 of [I-D.ietf-ippm-testplan-rfc2679] indicates that 124 Type-P-One-way-Delay-Inverse-Percentile statistic has been 125 ignored in both implementations, so it is a candidate for removal 126 or deprecation in RFC2679bis (this small discrepancy does not 127 affect candidacy for advancement) (see section 5.4, below). 129 3. The IETF has reached consensus on guidance for reporting metrics 130 in [RFC6703], and this memo should be referenced in RFC2679bis to 131 incorporate recent experience where appropriate (see the last 132 list item of section 3.6, section 3.8, and section 5 below). 134 4. There is currently one erratum with status "Held for document 135 update" for [RFC2679], and it appears this minor revision and 136 additional text should be incorporated in RFC2679bis (see section 137 5.1). 139 A small number of updates to the [RFC2679] text have been proposed 140 (by the current Editor) in the text below, principally to reference 141 key IPPM RFCs that were approved after [RFC2679]. 143 2. Introduction 145 This memo defines a metric for one-way delay of packets across 146 Internet paths. It builds on notions introduced and discussed in the 147 IPPM Framework document, RFC 2330 [1]; the reader is assumed to be 148 familiar with that document. 150 This memo is intended to be parallel in structure to a companion 151 document for Packet Loss ("A One-way Packet Loss Metric for IPPM") 152 [2]. 154 Although RFC 2119 was written with protocols in mind, the key words 155 are used in this document for similar reasons. They are used to 156 ensure the results of measurements from two different implementations 157 are comparable, and to note instances when an implementation could 158 perturb the network. 160 The structure of the memo is as follows: 162 + A 'singleton' analytic metric, called Type-P-One-way-Delay, will be 163 introduced to measure a single observation of one-way delay. 165 + Using this singleton metric, a 'sample', called Type-P-One-way- 166 Delay-Poisson-Stream, will be introduced to measure a sequence of 167 singleton delays measured at times taken from a Poisson process. 169 + Using this sample, several 'statistics' of the sample will be 170 defined and discussed. This progression from singleton to sample to 171 statistics, with clear separation among them, is important. 173 Whenever a technical term from the IPPM Framework document is first 174 used in this memo, it will be tagged with a trailing asterisk. For 175 example, "term*" indicates that "term" is defined in the Framework. 177 2.1. Motivation 179 One-way delay of a Type-P* packet from a source host* to a 180 destination host is useful for several reasons: 182 + Some applications do not perform well (or at all) if end-to-end 183 delay between hosts is large relative to some threshold value. 185 + Erratic variation in delay makes it difficult (or impossible) to 186 support many real-time applications. 188 + The larger the value of delay, the more difficult it is for 189 transport-layer protocols to sustain high bandwidths. 191 + The minimum value of this metric provides an indication of the 192 delay due only to propagation and transmission delay. 194 + The minimum value of this metric provides an indication of the 195 delay that will likely be experienced when the path* traversed is 196 lightly loaded. 198 + Values of this metric above the minimum provide an indication of 199 the congestion present in the path. 201 The measurement of one-way delay instead of round-trip delay is 202 motivated by the following factors: 204 + In today's Internet, the path from a source to a destination may be 205 different than the path from the destination back to the source 206 ("asymmetric paths"), such that different sequences of routers are 207 used for the forward and reverse paths. Therefore round-trip 208 measurements actually measure the performance of two distinct paths 209 together. Measuring each path independently highlights the 210 performance difference between the two paths which may traverse 211 different Internet service providers, and even radically different 212 types of networks (for example, research versus commodity networks, 213 or ATM versus packet-over-SONET). 215 + Even when the two paths are symmetric, they may have radically 216 different performance characteristics due to asymmetric queueing. 218 + Performance of an application may depend mostly on the performance 219 in one direction. For example, a file transfer using TCP may depend 220 more on the performance in the direction that data flows, rather than 221 the direction in which acknowledgements travel. 223 + In quality-of-service (QoS) enabled networks, provisioning in one 224 direction may be radically different than provisioning in the reverse 225 direction, and thus the QoS guarantees differ. Measuring the paths 226 independently allows the verification of both guarantees. 228 It is outside the scope of this document to say precisely how delay 229 metrics would be applied to specific problems. 231 2.2. General Issues Regarding Time 233 {Comment: the terminology below differs from that defined by ITU-T 234 documents (e.g., G.810, "Definitions and terminology for 235 synchronization networks" and I.356, "B-ISDN ATM layer cell transfer 236 performance"), but is consistent with the IPPM Framework document. 237 In general, these differences derive from the different backgrounds; 238 the ITU-T documents historically have a telephony origin, while the 239 authors of this document (and the Framework) have a computer systems 240 background. Although the terms defined below have no direct 241 equivalent in the ITU-T definitions, after our definitions we will 242 provide a rough mapping. However, note one potential confusion: our 243 definition of "clock" is the computer operating systems definition 244 denoting a time-of-day clock, while the ITU-T definition of clock 245 denotes a frequency reference.} 247 Whenever a time (i.e., a moment in history) is mentioned here, it is 248 understood to be measured in seconds (and fractions) relative to UTC. 250 As described more fully in the Framework document, there are four 251 distinct, but related notions of clock uncertainty: 253 synchronization* 255 measures the extent to which two clocks agree on what time it is. 256 For example, the clock on one host might be 5.4 msec ahead of the 257 clock on a second host. {Comment: A rough ITU-T equivalent is "time 258 error".} 260 accuracy* 262 measures the extent to which a given clock agrees with UTC. For 263 example, the clock on a host might be 27.1 msec behind UTC. {Comment: 264 A rough ITU-T equivalent is "time error from UTC".} 266 resolution* 268 measures the precision of a given clock. For example, the clock on 269 an old Unix host might tick only once every 10 msec, and thus have a 270 resolution of only 10 msec. {Comment: A very rough ITU-T equivalent 271 is "sampling period".} 273 skew* 275 measures the change of accuracy, or of synchronization, with time. 276 For example, the clock on a given host might gain 1.3 msec per hour 277 and thus be 27.1 msec behind UTC at one time and only 25.8 msec an 278 hour later. In this case, we say that the clock of the given host 279 has a skew of 1.3 msec per hour relative to UTC, which threatens 280 accuracy. We might also speak of the skew of one clock relative to 281 another clock, which threatens synchronization. {Comment: A rough 282 ITU-T equivalent is "time drift".} 284 3. A Singleton Definition for One-way Delay 286 3.1. Metric Name: 288 Type-P-One-way-Delay 290 3.2. Metric Parameters: 292 + Src, the IP address of a host 294 + Dst, the IP address of a host 296 + T, a time 298 3.3. Metric Units: 300 The value of a Type-P-One-way-Delay is either a real number, or an 301 undefined (informally, infinite) number of seconds. 303 3.4. Definition: 305 For a real number dT, >>the *Type-P-One-way-Delay* from Src to Dst at 306 T is dT<< means that Src sent the first bit of a Type-P packet to Dst 307 at wire-time* T and that Dst received the last bit of that packet at 308 wire-time T+dT. 310 >>The *Type-P-One-way-Delay* from Src to Dst at T is undefined 311 (informally, infinite)<< means that Src sent the first bit of a 312 Type-P packet to Dst at wire-time T and that Dst did not receive that 313 packet. 315 Suggestions for what to report along with metric values appear in 316 Section 3.8 after a discussion of the metric, methodologies for 317 measuring the metric, and error analysis. 319 3.5. Discussion: 321 Type-P-One-way-Delay is a relatively simple analytic metric, and one 322 that we believe will afford effective methods of measurement. 324 The following issues are likely to come up in practice: 326 + Real delay values will be positive. Therefore, it does not make 327 sense to report a negative value as a real delay. However, an 328 individual zero or negative delay value might be useful as part of a 329 stream when trying to discover a distribution of a stream of delay 330 values. 332 + Since delay values will often be as low as the 100 usec to 10 msec 333 range, it will be important for Src and Dst to synchronize very 334 closely. GPS systems afford one way to achieve synchronization to 335 within several 10s of usec. Ordinary application of NTP may allow 336 synchronization to within several msec, but this depends on the 337 stability and symmetry of delay properties among those NTP agents 338 used, and this delay is what we are trying to measure. A combination 339 of some GPS-based NTP servers and a conservatively designed and 340 deployed set of other NTP servers should yield good results, but this 341 is yet to be tested. 343 + A given methodology will have to include a way to determine whether 344 a delay value is infinite or whether it is merely very large (and the 345 packet is yet to arrive at Dst). As noted by Mahdavi and Paxson [4], 346 simple upper bounds (such as the 255 seconds theoretical upper bound 347 on the lifetimes of IP packets [5]) could be used, but good 348 engineering, including an understanding of packet lifetimes, will be 349 needed in practice. {Comment: Note that, for many applications of 350 these metrics, the harm in treating a large delay as infinite might 351 be zero or very small. A TCP data packet, for example, that arrives 352 only after several multiples of the RTT may as well have been lost.} 354 + If the packet is duplicated along the path (or paths) so that 355 multiple non-corrupt copies arrive at the destination, then the 356 packet is counted as received, and the first copy to arrive 357 determines the packet's one-way delay. 359 + If the packet is fragmented and if, for whatever reason, reassembly 360 does not occur, then the packet will be deemed lost. 362 3.6. Methodologies: 364 As with other Type-P-* metrics, the detailed methodology will depend 365 on the Type-P (e.g., protocol number, UDP/TCP port number, size, 366 precedence). 368 Generally, for a given Type-P, the methodology would proceed as 369 follows: 371 + Arrange that Src and Dst are synchronized; that is, that they have 372 clocks that are very closely synchronized with each other and each 373 fairly close to the actual time. 375 + At the Src host, select Src and Dst IP addresses, and form a test 376 packet of Type-P with these addresses. Any 'padding' portion of the 377 packet needed only to make the test packet a given size should be 378 filled with randomized bits to avoid a situation in which the 379 measured delay is lower than it would otherwise be due to compression 380 techniques along the path. 382 + At the Dst host, arrange to receive the packet. 384 + At the Src host, place a timestamp in the prepared Type-P packet, 385 and send it towards Dst. 387 + If the packet arrives within a reasonable period of time, take a 388 timestamp as soon as possible upon the receipt of the packet. By 389 subtracting the two timestamps, an estimate of one-way delay can be 390 computed. Error analysis of a given implementation of the method 391 must take into account the closeness of synchronization between Src 392 and Dst. If the delay between Src's timestamp and the actual sending 393 of the packet is known, then the estimate could be adjusted by 394 subtracting this amount; uncertainty in this value must be taken into 395 account in error analysis. Similarly, if the delay between the 396 actual receipt of the packet and Dst's timestamp is known, then the 397 estimate could be adjusted by subtracting this amount; uncertainty in 398 this value must be taken into account in error analysis. See the 399 next section, "Errors and Uncertainties", for a more detailed 400 discussion. 402 + If the packet fails to arrive within a reasonable period of time, 403 the one-way delay is taken to be undefined (informally, infinite). 404 Note that the threshold of 'reasonable' is a parameter of the 405 methodology. These points are examined in detail in [RFC6703], 406 including analysis preferences to assign undefined delay to packets 407 that fail to arrive with the difficulties emerging from the informal 408 "infinite delay" assignment, and an estimation of an upper bound on 409 waiting time for packets in transit. Further, enforcing a specific 410 constant waiting time on stored singletons of one-way delay is 411 compliant with this specification and may allow the results to serve 412 more than one reporting audience. 414 Issues such as the packet format, the means by which Dst knows when 415 to expect the test packet, and the means by which Src and Dst are 416 synchronized are outside the scope of this document. {Comment: We 417 plan to document elsewhere our own work in describing such more 418 detailed implementation techniques and we encourage others to as 419 well.} 421 3.7. Errors and Uncertainties: 423 The description of any specific measurement method should include an 424 accounting and analysis of various sources of error or uncertainty. 425 The Framework document provides general guidance on this point, but 426 we note here the following specifics related to delay metrics: 428 + Errors or uncertainties due to uncertainties in the clocks of the 429 Src and Dst hosts. 431 + Errors or uncertainties due to the difference between 'wire time' 432 and 'host time'. 434 In addition, the loss threshold may affect the results. Each of 435 these are discussed in more detail below, along with a section 436 ("Calibration") on accounting for these errors and uncertainties. 438 3.7.1. Errors or uncertainties related to Clocks 440 The uncertainty in a measurement of one-way delay is related, in 441 part, to uncertainties in the clocks of the Src and Dst hosts. In 442 the following, we refer to the clock used to measure when the packet 443 was sent from Src as the source clock, we refer to the clock used to 444 measure when the packet was received by Dst as the destination clock, 445 we refer to the observed time when the packet was sent by the source 446 clock as Tsource, and the observed time when the packet was received 447 by the destination clock as Tdest. Alluding to the notions of 448 synchronization, accuracy, resolution, and skew mentioned in the 449 Introduction, we note the following: 451 + Any error in the synchronization between the source clock and the 452 destination clock will contribute to error in the delay measurement. 453 We say that the source clock and the destination clock have a 454 synchronization error of Tsynch if the source clock is Tsynch ahead 455 of the destination clock. Thus, if we know the value of Tsynch 456 exactly, we could correct for clock synchronization by adding Tsynch 457 to the uncorrected value of Tdest-Tsource. 459 + The accuracy of a clock is important only in identifying the time 460 at which a given delay was measured. Accuracy, per se, has no 461 importance to the accuracy of the measurement of delay. When 462 computing delays, we are interested only in the differences between 463 clock values, not the values themselves. 465 + The resolution of a clock adds to uncertainty about any time 466 measured with it. Thus, if the source clock has a resolution of 10 467 msec, then this adds 10 msec of uncertainty to any time value 468 measured with it. We will denote the resolution of the source clock 469 and the destination clock as Rsource and Rdest, respectively. 471 + The skew of a clock is not so much an additional issue as it is a 472 realization of the fact that Tsynch is itself a function of time. 473 Thus, if we attempt to measure or to bound Tsynch, this needs to be 474 done periodically. Over some periods of time, this function can be 475 approximated as a linear function plus some higher order terms; in 476 these cases, one option is to use knowledge of the linear component 477 to correct the clock. Using this correction, the residual Tsynch is 478 made smaller, but remains a source of uncertainty that must be 479 accounted for. We use the function Esynch(t) to denote an upper 480 bound on the uncertainty in synchronization. Thus, |Tsynch(t)| <= 481 Esynch(t). 483 Taking these items together, we note that naive computation Tdest- 484 Tsource will be off by Tsynch(t) +/- (Rsource + Rdest). Using the 485 notion of Esynch(t), we note that these clock-related problems 486 introduce a total uncertainty of Esynch(t)+ Rsource + Rdest. This 487 estimate of total clock-related uncertainty should be included in the 488 error/uncertainty analysis of any measurement implementation. 490 3.7.2. Errors or uncertainties related to Wire-time vs Host-time 492 As we have defined one-way delay, we would like to measure the time 493 between when the test packet leaves the network interface of Src and 494 when it (completely) arrives at the network interface of Dst, and we 495 refer to these as "wire times." If the timings are themselves 496 performed by software on Src and Dst, however, then this software can 497 only directly measure the time between when Src grabs a timestamp 498 just prior to sending the test packet and when Dst grabs a timestamp 499 just after having received the test packet, and we refer to these two 500 points as "host times". 502 To the extent that the difference between wire time and host time is 503 accurately known, this knowledge can be used to correct for host time 504 measurements and the corrected value more accurately estimates the 505 desired (wire time) metric. 507 To the extent, however, that the difference between wire time and 508 host time is uncertain, this uncertainty must be accounted for in an 509 analysis of a given measurement method. We denote by Hsource an 510 upper bound on the uncertainty in the difference between wire time 511 and host time on the Src host, and similarly define Hdest for the Dst 512 host. We then note that these problems introduce a total uncertainty 513 of Hsource+Hdest. This estimate of total wire-vs-host uncertainty 514 should be included in the error/uncertainty analysis of any 515 measurement implementation. 517 3.7.3. Calibration 519 Generally, the measured values can be decomposed as follows: 521 measured value = true value + systematic error + random error 523 If the systematic error (the constant bias in measured values) can be 524 determined, it can be compensated for in the reported results. 526 reported value = measured value - systematic error 528 therefore 530 reported value = true value + random error 532 The goal of calibration is to determine the systematic and random 533 error generated by the instruments themselves in as much detail as 534 possible. At a minimum, a bound ("e") should be found such that the 535 reported value is in the range (true value - e) to (true value + e) 536 at least 95 percent of the time. We call "e" the calibration error 537 for the measurements. It represents the degree to which the values 538 produced by the measurement instrument are repeatable; that is, how 539 closely an actual delay of 30 ms is reported as 30 ms. {Comment: 95 540 percent was chosen because (1) some confidence level is desirable to 541 be able to remove outliers, which will be found in measuring any 542 physical property; (2) a particular confidence level should be 543 specified so that the results of independent implementations can be 544 compared; and (3) even with a prototype user-level implementation, 545 95% was loose enough to exclude outliers.} 547 From the discussion in the previous two sections, the error in 548 measurements could be bounded by determining all the individual 549 uncertainties, and adding them together to form 551 Esynch(t) + Rsource + Rdest + Hsource + Hdest. 553 However, reasonable bounds on both the clock-related uncertainty 554 captured by the first three terms and the host-related uncertainty 555 captured by the last two terms should be possible by careful design 556 techniques and calibrating the instruments using a known, isolated, 557 network in a lab. 559 For example, the clock-related uncertainties are greatly reduced 560 through the use of a GPS time source. The sum of Esynch(t) + Rsource 561 + Rdest is small, and is also bounded for the duration of the 562 measurement because of the global time source. 564 The host-related uncertainties, Hsource + Hdest, could be bounded by 565 connecting two instruments back-to-back with a high-speed serial link 566 or isolated LAN segment. In this case, repeated measurements are 567 measuring the same one-way delay. 569 If the test packets are small, such a network connection has a 570 minimal delay that may be approximated by zero. The measured delay 571 therefore contains only systematic and random error in the 572 instrumentation. The "average value" of repeated measurements is the 573 systematic error, and the variation is the random error. 575 One way to compute the systematic error, and the random error to a 576 95% confidence is to repeat the experiment many times - at least 577 hundreds of tests. The systematic error would then be the median. 578 The random error could then be found by removing the systematic error 579 from the measured values. The 95% confidence interval would be the 580 range from the 2.5th percentile to the 97.5th percentile of these 581 deviations from the true value. The calibration error "e" could then 582 be taken to be the largest absolute value of these two numbers, plus 583 the clock-related uncertainty. {Comment: as described, this bound is 584 relatively loose since the uncertainties are added, and the absolute 585 value of the largest deviation is used. As long as the resulting 586 value is not a significant fraction of the measured values, it is a 587 reasonable bound. If the resulting value is a significant fraction 588 of the measured values, then more exact methods will be needed to 589 compute the calibration error.} 591 Note that random error is a function of measurement load. For 592 example, if many paths will be measured by one instrument, this might 593 increase interrupts, process scheduling, and disk I/O (for example, 594 recording the measurements), all of which may increase the random 595 error in measured singletons. Therefore, in addition to minimal load 596 measurements to find the systematic error, calibration measurements 597 should be performed with the same measurement load that the 598 instruments will see in the field. 600 We wish to reiterate that this statistical treatment refers to the 601 calibration of the instrument; it is used to "calibrate the meter 602 stick" and say how well the meter stick reflects reality. 604 In addition to calibrating the instruments for finite one-way delay, 605 two checks should be made to ensure that packets reported as losses 606 were really lost. First, the threshold for loss should be verified. 607 In particular, ensure the "reasonable" threshold is reasonable: that 608 it is very unlikely a packet will arrive after the threshold value, 609 and therefore the number of packets lost over an interval is not 610 sensitive to the error bound on measurements. Second, consider the 611 possibility that a packet arrives at the network interface, but is 612 lost due to congestion on that interface or to other resource 613 exhaustion (e.g. buffers) in the instrument. 615 3.8. Reporting the metric: 617 The calibration and context in which the metric is measured MUST be 618 carefully considered, and SHOULD always be reported along with metric 619 results. We now present four items to consider: the Type-P of test 620 packets, the threshold of infinite delay (if any), error calibration, 621 and the path traversed by the test packets. This list is not 622 exhaustive; any additional information that could be useful in 623 interpreting applications of the metrics should also be reported (see 624 [RFC6703] for extensive discussion of reporting considerations for 625 different audiences). 627 3.8.1. Type-P 629 As noted in the Framework document [1], the value of the metric may 630 depend on the type of IP packets used to make the measurement, or 631 "type-P". The value of Type-P-One-way-Delay could change if the 632 protocol (UDP or TCP), port number, size, or arrangement for special 633 treatment (e.g., IP precedence or RSVP) changes. The exact Type-P 634 used to make the measurements MUST be accurately reported. 636 3.8.2. Loss Threshold 638 In addition, the threshold (or methodology to distinguish) between a 639 large finite delay and loss MUST be reported. 641 3.8.3. Calibration Results 643 + If the systematic error can be determined, it SHOULD be removed 644 from the measured values. 646 + You SHOULD also report the calibration error, e, such that the true 647 value is the reported value plus or minus e, with 95% confidence (see 648 the last section.) 650 + If possible, the conditions under which a test packet with finite 651 delay is reported as lost due to resource exhaustion on the 652 measurement instrument SHOULD be reported. 654 3.8.4. Path 656 Finally, the path traversed by the packet SHOULD be reported, if 657 possible. In general it is impractical to know the precise path a 658 given packet takes through the network. The precise path may be 659 known for certain Type-P on short or stable paths. If Type-P 660 includes the record route (or loose-source route) option in the IP 661 header, and the path is short enough, and all routers* on the path 662 support record (or loose-source) route, then the path will be 663 precisely recorded. This is impractical because the route must be 664 short enough, many routers do not support (or are not configured for) 665 record route, and use of this feature would often artificially worsen 666 the performance observed by removing the packet from common-case 667 processing. However, partial information is still valuable context. 668 For example, if a host can choose between two links* (and hence two 669 separate routes from Src to Dst), then the initial link used is 670 valuable context. {Comment: For example, with Merit's NetNow setup, a 671 Src on one NAP can reach a Dst on another NAP by either of several 672 different backbone networks.} 674 4. A Definition for Samples of One-way Delay 676 Given the singleton metric Type-P-One-way-Delay, we now define one 677 particular sample of such singletons. The idea of the sample is to 678 select a particular binding of the parameters Src, Dst, and Type-P, 679 then define a sample of values of parameter T. The means for defining 680 the values of T is to select a beginning time T0, a final time Tf, 681 and an average rate lambda, then define a pseudo-random Poisson 682 process of rate lambda, whose values fall between T0 and Tf. The 683 time interval between successive values of T will then average 684 1/lambda. 686 {Comment: Note that Poisson sampling is only one way of defining a 687 sample. Poisson has the advantage of limiting bias, but other 688 methods of sampling might be appropriate for different situations. 689 We encourage others who find such appropriate cases to use this 690 general framework and submit their sampling method for 691 standardization.} 693 >>> Editor proposal: Add ref to RFC 3432 Periodic sampling above. 695 4.1. Metric Name: 697 Type-P-One-way-Delay-Poisson-Stream 699 4.2. Metric Parameters: 701 + Src, the IP address of a host 703 + Dst, the IP address of a host 705 + T0, a time 707 + Tf, a time 709 + lambda, a rate in reciprocal seconds 711 4.3. Metric Units: 713 A sequence of pairs; the elements of each pair are: 715 + T, a time, and 717 + dT, either a real number or an undefined number of seconds. 719 The values of T in the sequence are monotonic increasing. Note that 720 T would be a valid parameter to Type-P-One-way-Delay, and that dT 721 would be a valid value of Type-P-One-way-Delay. 723 4.4. Definition: 725 Given T0, Tf, and lambda, we compute a pseudo-random Poisson process 726 beginning at or before T0, with average arrival rate lambda, and 727 ending at or after Tf. Those time values greater than or equal to T0 728 and less than or equal to Tf are then selected. At each of the times 729 in this process, we obtain the value of Type-P-One-way-Delay at this 730 time. The value of the sample is the sequence made up of the 731 resulting pairs. If there are no such pairs, the 732 sequence is of length zero and the sample is said to be empty. 734 4.5. Discussion: 736 The reader should be familiar with the in-depth discussion of Poisson 737 sampling in the Framework document [1], which includes methods to 738 compute and verify the pseudo-random Poisson process. 740 We specifically do not constrain the value of lambda, except to note 741 the extremes. If the rate is too large, then the measurement traffic 742 will perturb the network, and itself cause congestion. If the rate 743 is too small, then you might not capture interesting network 744 behavior. {Comment: We expect to document our experiences with, and 745 suggestions for, lambda elsewhere, culminating in a "best current 746 practices" document.} 748 Since a pseudo-random number sequence is employed, the sequence of 749 times, and hence the value of the sample, is not fully specified. 750 Pseudo-random number generators of good quality will be needed to 751 achieve the desired qualities. 753 The sample is defined in terms of a Poisson process both to avoid the 754 effects of self-synchronization and also capture a sample that is 755 statistically as unbiased as possible. {Comment: there is, of course, 756 no claim that real Internet traffic arrives according to a Poisson 757 arrival process.} The Poisson process is used to schedule the delay 758 measurements. The test packets will generally not arrive at Dst 759 according to a Poisson distribution, since they are influenced by the 760 network. 762 All the singleton Type-P-One-way-Delay metrics in the sequence will 763 have the same values of Src, Dst, and Type-P. 765 Note also that, given one sample that runs from T0 to Tf, and given 766 new time values T0' and Tf' such that T0 <= T0' <= Tf' <= Tf, the 767 subsequence of the given sample whose time values fall between T0' 768 and Tf' are also a valid Type-P-One-way-Delay-Poisson-Stream sample. 770 4.6. Methodologies: 772 The methodologies follow directly from: 774 + the selection of specific times, using the specified Poisson 775 arrival process, and 776 + the methodologies discussion already given for the singleton Type- 777 P-One-way-Delay metric. 779 Care must, of course, be given to correctly handle out-of-order 780 arrival of test packets; it is possible that the Src could send one 781 test packet at TS[i], then send a second one (later) at TS[i+1], 782 while the Dst could receive the second test packet at TR[i+1], and 783 then receive the first one (later) at TR[i]. 785 >>> Editor proposal: Add ref to RFC 4737 Reordering metric above. 787 4.7. Errors and Uncertainties: 789 In addition to sources of errors and uncertainties associated with 790 methods employed to measure the singleton values that make up the 791 sample, care must be given to analyze the accuracy of the Poisson 792 process with respect to the wire-times of the sending of the test 793 packets. Problems with this process could be caused by several 794 things, including problems with the pseudo-random number techniques 795 used to generate the Poisson arrival process, or with jitter in the 796 value of Hsource (mentioned above as uncertainty in the singleton 797 delay metric). The Framework document shows how to use the Anderson- 798 Darling test to verify the accuracy of a Poisson process over small 799 time frames. {Comment: The goal is to ensure that test packets are 800 sent "close enough" to a Poisson schedule, and avoid periodic 801 behavior.} 803 4.8. Reporting the metric: 805 You MUST report the calibration and context for the underlying 806 singletons along with the stream. (See "Reporting the metric" for 807 Type-P-One-way-Delay.) 809 5. Some Statistics Definitions for One-way Delay 811 Given the sample metric Type-P-One-way-Delay-Poisson-Stream, we now 812 offer several statistics of that sample. These statistics are 813 offered mostly to be illustrative of what could be done. See 814 [RFC6703] for additional discussion of statistics that are relevant 815 to different audiences. 817 5.1. Type-P-One-way-Delay-Percentile 819 Given a Type-P-One-way-Delay-Poisson-Stream and a percent X between 820 0% and 100%, the Xth percentile of all the dT values in the Stream. 821 In computing this percentile, undefined values are treated as 822 infinitely large. Note that this means that the percentile could 823 thus be undefined (informally, infinite). In addition, the Type-P- 824 One-way-Delay-Percentile is undefined if the sample is empty. 826 Example: suppose we take a sample and the results are: 828 Stream1 = < 830 832 834 836 838 840 > 842 Then the 50th percentile would be 110 msec, since 90 msec and 100 843 msec are smaller and 500 msec and 'undefined' are larger. See 844 Section 11.3 of [1] for computing percentiles. 846 Note that if the possibility that a packet with finite delay is 847 reported as lost is significant, then a high percentile (90th or 848 95th) might be reported as infinite instead of finite. 850 5.2. Type-P-One-way-Delay-Median 852 Given a Type-P-One-way-Delay-Poisson-Stream, the median of all the dT 853 values in the Stream. In computing the median, undefined values are 854 treated as infinitely large. As with Type-P-One-way-Delay- 855 Percentile, Type-P-One-way-Delay-Median is undefined if the sample is 856 empty. 858 As noted in the Framework document, the median differs from the 50th 859 percentile only when the sample contains an even number of values, in 860 which case the mean of the two central values is used. 862 Example: suppose we take a sample and the results are: 864 Stream2 = < > 867 Then the median would be 105 msec, the mean of 100 msec and 110 msec, 868 the two central values. 870 5.3. Type-P-One-way-Delay-Minimum 872 Given a Type-P-One-way-Delay-Poisson-Stream, the minimum of all the 873 dT values in the Stream. In computing this, undefined values are 874 treated as infinitely large. Note that this means that the minimum 875 could thus be undefined (informally, infinite) if all the dT values 876 are undefined. In addition, the Type-P-One-way-Delay-Minimum is 877 undefined if the sample is empty. 879 In the above example, the minimum would be 90 msec. 881 5.4. Type-P-One-way-Delay-Inverse-Percentile 883 Note: This statistic is deprecated in this version of the memo 884 because of lack of use. 886 Given a Type-P-One-way-Delay-Poisson-Stream and a time duration 887 threshold, the fraction of all the dT values in the Stream less than 888 or equal to the threshold. The result could be as low as 0% (if all 889 the dT values exceed threshold) or as high as 100%. Type-P-One-way- 890 Delay-Inverse-Percentile is undefined if the sample is empty. 892 In the above example, the Inverse-Percentile of 103 msec would be 893 50%. 895 6. Security Considerations 897 Conducting Internet measurements raises both security and privacy 898 concerns. This memo does not specify an implementation of the 899 metrics, so it does not directly affect the security of the Internet 900 nor of applications which run on the Internet. However, 901 implementations of these metrics must be mindful of security and 902 privacy concerns. 904 There are two types of security concerns: potential harm caused by 905 the measurements, and potential harm to the measurements. The 906 measurements could cause harm because they are active, and inject 907 packets into the network. The measurement parameters MUST be 908 carefully selected so that the measurements inject trivial amounts of 909 additional traffic into the networks they measure. If they inject 910 "too much" traffic, they can skew the results of the measurement, and 911 in extreme cases cause congestion and denial of service. 913 The measurements themselves could be harmed by routers giving 914 measurement traffic a different priority than "normal" traffic, or by 915 an attacker injecting artificial measurement traffic. If routers can 916 recognize measurement traffic and treat it separately, the 917 measurements will not reflect actual user traffic. If an attacker 918 injects artificial traffic that is accepted as legitimate, the loss 919 rate will be artificially lowered. Therefore, the measurement 920 methodologies SHOULD include appropriate techniques to reduce the 921 probability measurement traffic can be distinguished from "normal" 922 traffic. Authentication techniques, such as digital signatures, may 923 be used where appropriate to guard against injected traffic attacks. 925 The privacy concerns of network measurement are limited by the active 926 measurements described in this memo. Unlike passive measurements, 927 there can be no release of existing user data. 929 7. IANA Considerations 931 This memo makes no requests of IANA. 933 8. Acknowledgements 935 Special thanks are due to Vern Paxson of Lawrence Berkeley Labs for 936 his helpful comments on issues of clock uncertainty and statistics. 937 Thanks also to Garry Couch, Will Leland, Andy Scherrer, Sean Shapira, 938 and Roland Wittig for several useful suggestions. 940 9. Refetrences (temporary) 942 [1] Paxson, V., Almes, G., Mahdavi, J. and M. Mathis, "Framework for 943 IP Performance Metrics", RFC 2330, May 1998. 945 [2] Almes, G., Kalidindi, S. and M. Zekauskas, "A One-way Packet Loss 946 Metric for IPPM", RFC 2680, September 1999. 948 [3] Mills, D., "Network Time Protocol (v3)", RFC 1305, April 1992. 950 [4] Mahdavi J. and V. Paxson, "IPPM Metrics for Measuring 951 Connectivity", RFC 2678, September 1999. 953 [5] Postel, J., "Internet Protocol", STD 5, RFC 791, September 1981. 955 [6] Bradner, S., "Key words for use in RFCs to Indicate Requirement 956 Levels", BCP 14, RFC 2119, March 1997. 958 [7] Bradner, S., "The Internet Standards Process -- Revision 3", BCP 959 9, RFC 2026, October 1996. 961 10. References 963 10.1. Normative References 965 [RFC2026] Bradner, S., "The Internet Standards Process -- Revision 966 3", BCP 9, RFC 2026, October 1996. 968 [RFC2119] Bradner, S., "Key words for use in RFCs to Indicate 969 Requirement Levels", BCP 14, RFC 2119, March 1997. 971 [RFC2330] Paxson, V., Almes, G., Mahdavi, J., and M. Mathis, 972 "Framework for IP Performance Metrics", RFC 2330, 973 May 1998. 975 [RFC2679] Almes, G., Kalidindi, S., and M. Zekauskas, "A One-way 976 Delay Metric for IPPM", RFC 2679, September 1999. 978 [RFC2680] Almes, G., Kalidindi, S., and M. Zekauskas, "A One-way 979 Packet Loss Metric for IPPM", RFC 2680, September 1999. 981 [RFC3432] Raisanen, V., Grotefeld, G., and A. Morton, "Network 982 performance measurement with periodic streams", RFC 3432, 983 November 2002. 985 [RFC4656] Shalunov, S., Teitelbaum, B., Karp, A., Boote, J., and M. 986 Zekauskas, "A One-way Active Measurement Protocol 987 (OWAMP)", RFC 4656, September 2006. 989 [RFC5357] Hedayat, K., Krzanowski, R., Morton, A., Yum, K., and J. 990 Babiarz, "A Two-Way Active Measurement Protocol (TWAMP)", 991 RFC 5357, October 2008. 993 [RFC5657] Dusseault, L. and R. Sparks, "Guidance on Interoperation 994 and Implementation Reports for Advancement to Draft 995 Standard", BCP 9, RFC 5657, September 2009. 997 [RFC5835] Morton, A. and S. Van den Berghe, "Framework for Metric 998 Composition", RFC 5835, April 2010. 1000 [RFC6049] Morton, A. and E. Stephan, "Spatial Composition of 1001 Metrics", RFC 6049, January 2011. 1003 [RFC6576] Geib, R., Morton, A., Fardid, R., and A. Steinmitz, "IP 1004 Performance Metrics (IPPM) Standard Advancement Testing", 1005 BCP 176, RFC 6576, March 2012. 1007 [RFC6703] Morton, A., Ramachandran, G., and G. Maguluri, "Reporting 1008 IP Network Performance Metrics: Different Points of View", 1009 RFC 6703, August 2012. 1011 10.2. Informative References 1013 [ADK] Scholz, F. and M. Stephens, "K-sample Anderson-Darling 1014 Tests of fit, for continuous and discrete cases", 1015 University of Washington, Technical Report No. 81, 1016 May 1986. 1018 [I-D.ietf-ippm-testplan-rfc2679] 1019 Ciavattone, L., Geib, R., Morton, A., and M. Wieser, "Test 1020 Plan and Results Supporting Advancement of RFC 2679 on the 1021 Standards Track", draft-ietf-ippm-testplan-rfc2679-03 1022 (work in progress), September 2012. 1024 [RFC3931] Lau, J., Townsley, M., and I. Goyret, "Layer Two Tunneling 1025 Protocol - Version 3 (L2TPv3)", RFC 3931, March 2005. 1027 Authors' Addresses 1029 Guy Almes 1030 Texas A&M 1032 Phone: 1033 Fax: 1034 Email: 1035 URI: 1037 Sunil Kalidindi 1038 Ixia 1040 Phone: 1041 Fax: 1042 Email: 1043 URI: 1045 Matt Zekauskas 1046 Internet2 1048 Phone: 1049 Fax: 1050 Email: matt@internet2.edu 1051 URI: 1053 Al Morton (editor) 1054 AT&T Labs 1055 200 Laurel Avenue South 1056 Middletown, NJ 07748 1057 USA 1059 Phone: +1 732 420 1571 1060 Fax: +1 732 368 1192 1061 Email: acmorton@att.com 1062 URI: http://home.comcast.net/~acmacm/