idnits 2.17.1 draft-ietf-ippm-npmps-06.txt: ** The Abstract section seems to be numbered Checking boilerplate required by RFC 5378 and the IETF Trust (see https://trustee.ietf.org/license-info): ---------------------------------------------------------------------------- ** Looks like you're using RFC 2026 boilerplate. This must be updated to follow RFC 3978/3979, as updated by RFC 4748. Checking nits according to https://www.ietf.org/id-info/1id-guidelines.txt: ---------------------------------------------------------------------------- ** Missing expiration date. The document expiration date should appear on the first and last page. ** The document seems to lack a 1id_guidelines paragraph about 6 months document validity -- however, there's a paragraph with a matching beginning. Boilerplate error? ** The document is more than 15 pages and seems to lack a Table of Contents. Checking nits according to https://www.ietf.org/id-info/checklist : ---------------------------------------------------------------------------- ** The document seems to lack separate sections for Informative/Normative References. All references will be assumed normative when checking for downward references. ** There are 14 instances of too long lines in the document, the longest one being 2 characters in excess of 72. Miscellaneous warnings: ---------------------------------------------------------------------------- == Line 285 has weird spacing: '... derive the s...' == Line 365 has weird spacing: '... to the clock...' == Line 689 has weird spacing: '...s point of pa...' == Line 690 has weird spacing: '...lose to the i...' == Line 693 has weird spacing: '... follow the c...' -- The document seems to lack a disclaimer for pre-RFC5378 work, but may have content which was first submitted before 10 November 2008. If you have contacted all the original authors and they are all willing to grant the BCP78 rights to the IETF Trust, then this is fine, and you can ignore this comment. If not, you may need to add the pre-RFC5378 disclaimer. (See the Legal Provisions document at https://trustee.ietf.org/license-info for more information.) -- Couldn't find a document date in the document -- date freshness check skipped. Checking references for intended status: Informational ---------------------------------------------------------------------------- -- Missing reference section? '1' on line 15 looks like a reference -- Missing reference section? '2' on line 56 looks like a reference -- Missing reference section? '3' on line 921 looks like a reference -- Missing reference section? '4' on line 666 looks like a reference -- Missing reference section? '5' on line 258 looks like a reference -- Missing reference section? 'I' on line 263 looks like a reference -- Missing reference section? '6' on line 594 looks like a reference -- Missing reference section? '7' on line 661 looks like a reference -- Missing reference section? '8' on line 667 looks like a reference Summary: 7 errors (**), 0 flaws (~~), 5 warnings (==), 11 comments (--). Run idnits with the --verbose option for more detailed information about the items above. -------------------------------------------------------------------------------- 2 IP Performance Measurement Working Group V.Raisanen 3 Internet Draft Nokia 4 Document: G.Grotefeld 5 Category: Informational Motorola 6 A.Morton 7 AT&T Labs 9 Network performance measurement with periodic streams 11 Status of this Memo 13 This document is an Internet-Draft and is in full conformance with 14 all provisions of Section 10 of RFC2026 [1]. 16 Internet-Drafts are working documents of the Internet Engineering 17 Task Force (IETF), its areas, and its working groups. Note that 18 other groups may also distribute working documents as Internet- 19 Drafts. Internet-Drafts are draft documents valid for a maximum of 20 six months and may be updated, replaced, or made obsolete by other 21 documents at any time. It is inappropriate to use Internet-Drafts as 22 reference material or to cite them other than as "work in progress." 24 The list of current Internet-Drafts can be accessed at 25 http://www.ietf.org/ietf/1id-abstracts.txt 27 The list of Internet-Draft Shadow Directories can be accessed at 28 http://www.ietf.org/shadow.html. 30 1. Abstract 32 This memo describes a periodic sampling method and relevant metrics 33 for assessing the performance of IP networks. First, the memo 34 motivates periodic sampling and addresses the question of its value 35 as an alternative to Poisson sampling described in RFC 2330. The 36 benefits include applicability to active and passive measurements, 37 simulation of constant bit rate (CBR) traffic (typical of multimedia 38 communication, or nearly CBR, as found with voice activity 39 detection), and several instances where analysis can be simplified. 40 The sampling method avoids predictability by mandating random start 41 times and finite length tests. Following descriptions of the 42 sampling method and sample metric parameters, measurement methods 43 and errors are discussed. Finally, we give additional information on 44 periodic measurements including security considerations. 46 2. Conventions used in this document 48 The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", 49 "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this 50 document are to be interpreted as described in RFC 2119 [2]. 51 Although RFC 2119 was written with protocols in mind, the key words 52 are used in this document for similar reasons. They are used to 53 ensure the results of measurements from two different 54 implementations are comparable, and to note instances when an 55 implementation could perturb the network. 57 3. Introduction 59 This memo describes a sampling method and performance metrics 60 relevant to certain applications of IP networks. The original driver 61 for this work was Quality of Service of interactive periodic streams 62 such as multimedia conferencing over IP, but the idea of periodic 63 sampling and measurement has wider applicability. Interactive 64 multimedia traffic is used as an example below to illustrate the 65 concept. 67 Transmitting equal size packets (or mostly same-size packets) 68 through a network at regular intervals simulates a constant bit-rate 69 (CBR), or nearly CBR multimedia bit stream. Hereafter, these packets 70 are called periodic streams. Cases of "mostly same-size packets" may 71 be found in applications that have multiple coding methods (e.g. 72 digitally coded comfort noise during silence gaps in speech). 74 In the following sections, a sampling methodology and metrics are 75 presented for periodic streams. The measurement results may be used 76 in derivative metrics such as average and maximum delays. The memo 77 seeks to formalize periodic stream measurements to achieve 78 comparable results between independent implementations. 80 3.1 Motivation 82 As noted in the IPPM framework RFC 2330 [3], a sample metric using 83 regularly spaced singleton tests has some limitations when 84 considered from a general measurement point of view: only part of 85 the network performance spectrum is sampled. However, some 86 applications also sample this limited performance spectrum and their 87 performance may be of critical interest. 89 Periodic sampling is useful for the following reasons: 91 * It is applicable to passive measurement, as well as active 92 measurement. 94 * An active measurement can be configured to match the 95 characteristics of media flows, and simplifies the estimation of 96 application performance. 98 * Measurements of many network impairments (e.g., delay variation, 99 consecutive loss, reordering) are sensitive to the sampling 100 frequency. When the impairments themselves are time-varying (and 101 the variations are somewhat rare, yet important), a constant 102 sampling frequency simplifies analysis. 104 * Frequency Domain analysis is simplified when the samples are 105 equally spaced. 107 Simulation of CBR flows with periodic streams encourages dense 108 sampling of network performance, since typical multimedia flows have 109 10 to 100 packets in each second. Dense sampling permits the 110 characterization of network phenomena with short duration. 112 4. Periodic Sampling Methodology 114 The Framework RFC [3] points out the following potential problems 115 with Periodic Sampling: 117 1. The performance sampled may be synchronized with some other 118 periodic behavior, or the samples may be anticipated and the results 119 manipulated. Unpredictable sampling is preferred. 121 2. Active measurements can cause congestion, and periodic sampling 122 might drive congestion-aware senders into a synchronized state, 123 producing atypical results. 125 Poisson sampling produces an unbiased sample for the various IP 126 performance metrics, yet there are situations where alternative 127 sampling methods are advantageous (as discussed under Motivation). 129 We can prescribe periodic sampling methods that address the problems 130 listed above. Predictability and some forms of synchronization can 131 be mitigated through the use of random start times and limited 132 stream duration over a test interval. The periodic sampling 133 parameters produce bias, and judicious selection can produce a known 134 bias of interest. The total traffic generated by this or any 135 sampling method should be limited to avoid adverse affects on non- 136 test traffic (packet size, packet rate, and sample duration and 137 frequency should all be considered). 139 The configuration parameters of periodic sampling are: 141 + T, the beginning of a time interval where a periodic sample is 142 desired. 143 + dT, the duration of the interval for allowed sample start times. 144 + T0, a time that MUST be selected at random from the interval [T, 145 T+dT] to start generating packets and taking measurements. 146 + Tf, a time, greater than T0, for stopping generation of packets 147 for a sample (Tf may be relative to T0 if desired). 148 + incT, the nominal duration of inter-packet interval, first bit to 149 first bit. 151 T0 may be drawn from a uniform distribution, or T0 = T + Unif(0,dT). 152 Other distributions may also be appropriate. Start times in 153 successive time intervals MUST use an independent value drawn from 154 the distribution. In passive measurement, the arrival of user media 155 flows may have sufficient randomness, or a randomized start time of 156 the measurement during a flow may be needed to meet this 157 requirement. 159 When a mix of packet sizes is desired, passive measurements usually 160 possess the sequence and statistics of sizes in actual use, while 161 active measurements would need to reproduce the intended 162 distribution of sizes. 164 5. Sample metrics for periodic streams 166 The sample metric presented here is similar to the sample metric 167 Type-P-One-way-Delay-Poisson-Stream presented in RFC 2679[4]. 168 Singletons defined in [3] and [4] are applicable here. 170 5.1 Metric name 172 Type-P-One-way-Delay-Periodic-Stream 174 5.2 Metric parameters 176 5.2.1 Global metric parameters 178 These parameters apply in all the sub-sections that follow (5.2.2, 179 5.2.3, and 5.2.4). 181 Parameters that each Singleton usually includes: 182 + Src, the IP address of a host 183 + Dst, the IP address of a host 184 + IPV, the IP version (IPv4/IPv6) used in the measurement 185 + dTloss, a time interval, the maximum waiting time for a packet 186 before declaring it lost. 187 + packet size p(j), the desired number of bytes in the Type-P 188 packet, where j is the size index. 190 Optional parameters: 191 + PktType, any additional qualifiers (transport address) 192 + Tcons, a time interval for consolidating parameters collected at 193 the measurement points. 195 While a number of applications will use one packet size (j = 1), 196 other applications may use packets of different sizes (j > 1). 197 Especially in cases of congestion, it may be useful to use packets 198 smaller than the maximum or predominant size of packets in the 199 periodic stream. 201 A topology where Src and Dst are separate from the measurement 202 points is assumed. 204 5.2.2 Parameters collected at the measurement point MP(Src) 206 Parameters that each Singleton usually includes: 208 + Tstamp(Src)[i], for each packet [i], the time of the packet as 209 measured at MP(Src) 211 Additional parameters: 212 + PktID(Src) [i], for each packet [i], a unique identification or 213 sequence number. 214 + PktSi(Src) [i], for each packet [i], the actual packet size. 216 Some applications may use packets of different sizes, either 217 because of application requirements or in response to IP 218 performance experienced. 220 5.2.3 Parameters collected at the measurement point MP(Dst) 222 + Tstamp(Dst)[i], for each packet [i], the time of the packet as 223 measured at MP(Dst) 224 + PktID(Dst) [i], for each packet [i], a unique identification or 225 sequence number. 226 + PktSi(Dst) [i], for each packet [i], the actual packet size. 228 Optional parameters: 229 + dTstop, a time interval, used to add to time Tf to determine when 230 to stop collecting metrics for a sample 231 + PktStatus [i], for each packet [i], the status of the packet 232 received. Possible status includes OK, packet header corrupt, 233 packet payload corrupt, duplicate, fragment. The criteria to 234 determine the status MUST be specified, if used. 236 5.2.4 Sample Metrics resulting from combining parameters at MP(Src) and 237 MP(Dst) 239 Using the parameters above, a delay singleton would be calculated as 240 follows: 241 + Delay [i], for each packet [i], the time interval 242 Delay[i] = Tstamp(Dst)[i] - Tstamp(Src)[i] 244 For the following conditions, it will not be possible to be able to 245 compute delay singletons: 247 Spurious: There will be no Tstamp(Src)[i] time 248 Not received: There will be no Tstamp (Dst) [i] 249 Corrupt packet header: There will be no Tstamp (Dst) [i] 250 Duplicate: Only the first non-corrupt copy of the packet 251 received at Dst should have Delay [i] computed. 253 A sample metric for average delay is as follows 255 AveDelay = (1/N)Sum(from i=1 to N, Delay[i]) 256 assuming all packets i= 1 though N have valid singletons. 258 A delay variation [5] singleton can also be computed: 260 + IPDV[i], for each packet [i] except the first one, delay 261 variation between successive packets would be calculated as 263 IPDV[I] = Delay[i] - Delay [i-1] 265 IPDV[i] may be negative, zero, or positive. Delay singletons for 266 packets i and i-1 must be calculable or IPDV[i] is undefined. 268 An example metric for the IPDV sample is the range: 270 RangeIPDV = max(IPDV[]) - min(IPDV[]) 272 5.3 High level description of the procedure to collect a sample 274 Beginning on or after time T0, Type-P packets are generated by Src 275 and sent to Dst until time Tf is reached with a nominal interval 276 between the first bit of successive packets of incT as measured at 277 MP(Src). incT may be nominal due to a number of reasons: variation 278 in packet generation at Src, clock issues (see section 5.6), etc. 279 MP(Src) records the parameters above only for packets with 280 timestamps between and including T0 and Tf having the required Src, 281 Dst, and any other qualifiers. MP (Dst) also records for packets 282 with time stamps between T0 and (Tf + dTstop). 284 Optionally at a time Tf + Tcons (but eventually in all cases), the 285 data from MP(Src) and MP(Dst) are consolidated to derive the sample 286 metric results. To prevent stopping data collection too soon, 287 dTcons should be greater than or equal to dTstop. Conversely, to 288 keep data collection reasonably efficient, dTstop should be some 289 reasonable time interval (seconds/minutes/hours), even if dTloss is 290 infinite or extremely long. 292 5.4 Discussion 294 This sampling methodology is intended to quantify the delays and the 295 delay variation as experienced by multimedia streams of an 296 application. Due to the definitions of these metrics, also packet 297 loss status is recorded. The nominal interval between packets 298 assesses network performance variations on a specific time scale. 300 There are a number of factors that should be taken into account when 301 collecting a sample metric of Type-P-One-way-Delay-Periodic-Stream. 303 + The interval T0 to Tf should be specified to cover a long enough 304 time interval to represent a reasonable use of the application under 305 test, yet not excessively long in the same context(e.g. phone calls 306 last longer than 100ms, but less than one week). 308 + The nominal interval between packets (incT) and the packet 309 size(s) (p(j)) should not define an equivalent bit rate that exceeds 310 the capacity of the egress port of Src, the ingress port of Dst, 311 or the capacity of the intervening network(s), if known. There may 312 be exceptional cases to test the response of the application to 313 overload conditions in the transport networks, but these cases 314 should be strictly controlled. 316 + Real delay values will be positive. Therefore, it does not make 317 sense to report a negative value as a real delay. However, an 318 individual zero or negative delay value might be useful as part of 319 a stream when trying to discover a distribution of the delay errors. 321 + Depending on measurement topology, delay values may be as low as 322 100 usec to 10 msec, whereby it may be important for Src and Dst to 323 synchronize very closely. GPS systems afford one way to achieve 324 synchronization to within several 10s of usec. Ordinary application 325 of NTP may allow synchronization to within several msec, but this 326 depends on the stability and symmetry of delay properties among the 327 NTP agents used, and this delay is what we are trying to measure. 329 + A given methodology will have to include a way to determine 330 whether packet was lost or whether delay is merely very large (and 331 the packet is yet to arrive at Dst). The global metric parameter 332 dTloss defines a time interval such that delays larger than dTloss 333 are interpreted as losses. {Comment: For many applications, the 334 treatment a large delay as infinite/loss will be inconsequential. A 335 TCP data packet, for example, that arrives only after several 336 multiples of the usual RTT may as well have been lost.} 338 5.5 Additional Methodology Aspects 340 As with other Type-P-* metrics, the detailed methodology will depend 341 on the Type-P (e.g., protocol number, UDP/TCP port number, size, 342 precedence). 344 5.6 Errors and uncertainties 346 The description of any specific measurement method should include an 347 accounting and analysis of various sources of error or uncertainty. 348 The Framework RFC [3] provides general guidance on this point, but 349 we note here the following specifics related to periodic streams and 350 delay metrics: 352 + Error due to variation of incT. The reasons for this can be 353 uneven process scheduling, possibly due to CPU load. 355 + Errors or uncertainties due to uncertainties in the clocks of the 356 MP(Src) and MP(Dst) measurement points. 358 + Errors or uncertainties due to the difference between 'wire time' 359 and 'host time'. 361 5.6.1. Errors or uncertainties related to Clocks 362 The uncertainty in a measurement of one-way delay is related, in 363 part, to uncertainties in the clocks of MP(Src) and MP(Dst). In the 364 following, we refer to the clock used to measure when the packet was 365 measured at MP(Src) as the MP(Src) clock and we refer to the clock 366 used to measure when the packet was received at MP(Dst) as the 367 MP(Dst) clock. Alluding to the notions of synchronization, 368 accuracy, resolution, and skew, we note the following: 370 + Any error in the synchronization between the MP(Src) clock and 371 the MP(Dst) clock will contribute to error in the delay measurement. 372 We say that the MP(Src) clock and the MP(Dst) clock have a 373 synchronization error of Tsynch if the MP(Src) clock is Tsynch ahead 374 of the MP(Dst) clock. Thus, if we know the value of Tsynch exactly, 375 we could correct for clock synchronization by adding Tsynch to the 376 uncorrected value of Tstamp(Dst)[i] - Tstamp(Src) [i]. 378 + The resolution of a clock adds to uncertainty about any time 379 measured with it. Thus, if the MP(Src) clock has a resolution of 380 10 msec, then this adds 10 msec of uncertainty to any time value 381 measured with it. We will denote the resolution of the source 382 clock and the MP(Dst) clock as ResMP(Src) and ResMP(Dst), 383 respectively. 385 + The skew of a clock is not so much an additional issue as it is a 386 realization of the fact that Tsynch is itself a function of time. 387 Thus, if we attempt to measure or to bound Tsynch, this needs to 388 be done periodically. Over some periods of time, this function can 389 be approximated as a linear function plus some higher order terms; 390 in these cases, one option is to use knowledge of the linear 391 component to correct the clock. Using this correction, the residual 392 Tsynch is made smaller, but remains a source of uncertainty that 393 must be accounted for. We use the function Esynch(t) to denote an 394 upper bound on the uncertainty in synchronization. Thus, 395 |Tsynch(t)| <= Esynch(t). 397 Taking these items together, we note that naive computation 398 Tstamp(Dst)[i] - Tstamp(Src) [i] will be off by Tsynch(t) +/- 399 (ResMP(SRc) + ResMP(Dst)). Using the notion of Esynch(t), we note 400 that these clock-related problems introduce a total uncertainty of 401 Esynch(t)+ Rsource + Rdest. This estimate of total clock-related 402 uncertainty should be included in the error/uncertainty analysis of 403 any measurement implementation. 405 5.6.2. Errors or uncertainties related to Wire-time vs Host-time 407 We would like to measure the time between when a packet is measured 408 and time-stamped at MP(Src) and when it arrives and is time-stamped 409 at MP(Dst) and we refer to these as "wire times." If timestamps are 410 applied by software on Src and Dst, however, then this software can 411 only directly measure the time between when Src generates the packet 412 just prior to sending the test packet and when Dst has started to 413 process the packet after having received the test packet, and we 414 refer to these two points as "host times". 416 To the extent that the difference between wire time and host time is 417 accurately known, this knowledge can be used to correct for wire 418 time measurements and the corrected value more accurately estimates 419 the desired (host time) metric, and visa-versa. 421 To the extent, however, that the difference between wire time and 422 host time is uncertain, this uncertainty must be accounted for in an 423 analysis of a given measurement method. We denote by Hsource an 424 upper bound on the uncertainty in the difference between wire time 425 of MP(Src) and host time on the Src host, and similarly define Hdest 426 for the difference between the host time on the Dst host and the 427 wire time of MP(Dst). We then note that these problems introduce a 428 total uncertainty of Hsource+Hdest. This estimate of total wire-vs- 429 host uncertainty should be included in the error/uncertainty 430 analysis of any measurement implementation. 432 5.6.3. Calibration 434 Generally, the measured values can be decomposed as follows: 436 measured value = true value + systematic error + random error 438 If the systematic error (the constant bias in measured values) can 439 be determined, it can be compensated for in the reported results. 441 reported value = measured value - systematic error 443 therefore 445 reported value = true value + random error 447 The goal of calibration is to determine the systematic and random 448 error generated by the instruments themselves in as much detail as 449 possible. At a minimum, a bound ("e") should be found such that the 450 reported value is in the range (true value - e) to (true value + e) 451 at least 95 percent of the time. We call "e" the calibration error 452 for the measurements. It represents the degree to which the values 453 produced by the measurement instrument are repeatable; that is, how 454 closely an actual delay of 30 ms is reported as 30 ms. 455 {Comment: 95 percent was chosen due to reasons discussed in [4], 456 briefly summarized as (1) some confidence level is desirable to be 457 able to remove outliers, which will be found in measuring any 458 physical property; (2) a particular confidence level should be 459 specified so that the results of independent implementations can be 460 compared.} 462 From the discussion in the previous two sections, the error in 463 measurements could be bounded by determining all the individual 464 uncertainties, and adding them together to form 465 Esynch(t) + ResMP(Src) + ResMP(Dst) + Hsource + Hdest. 467 However, reasonable bounds on both the clock-related uncertainty 468 captured by the first three terms and the host-related uncertainty 469 captured by the last two terms should be possible by careful design 470 techniques and calibrating the instruments using a known, isolated, 471 network in a lab. 473 For example, the clock-related uncertainties are greatly reduced 474 through the use of a GPS time source. The sum of Esynch(t) + 475 ResMP(Src) + ResMP(Dst) is small, and is also bounded for the 476 duration of the measurement because of the global time source. 477 The host-related uncertainties, Hsource + Hdest, could be bounded by 478 connecting two instruments back-to-back with a high-speed serial 479 link or isolated LAN segment. In this case, repeated measurements 480 are measuring the same one-way delay. 482 If the test packets are small, such a network connection has a 483 minimal delay that may be approximated by zero. The measured delay 484 therefore contains only systematic and random error in the 485 instrumentation. The "average value" of repeated measurements is 486 the systematic error, and the variation is the random error. 487 One way to compute the systematic error, and the random error to a 488 95% confidence is to repeat the experiment many times - at least 489 hundreds of tests. The systematic error would then be the median. 490 The random error could then be found by removing the systematic 491 error from the measured values. The 95% confidence interval would 492 be the range from the 2.5th percentile to the 97.5th percentile of 493 these deviations from the true value. The calibration error "e" 494 could then be taken to be the largest absolute value of these two 495 numbers, plus the clock-related uncertainty. {Comment: as 496 described, this bound is relatively loose since the uncertainties 497 are added, and the absolute value of the largest deviation is used. 498 As long as the resulting value is not a significant fraction of the 499 measured values, it is a reasonable bound. If the resulting value 500 is a significant fraction of the measured values, then more exact 501 methods will be needed to compute the calibration error.} 503 Note that random error is a function of measurement load. For 504 example, if many paths will be measured by one instrument, this 505 might increase interrupts, process scheduling, and disk I/O (for 506 example, recording the measurements), all of which may increase the 507 random error in measured singletons. Therefore, in addition to 508 minimal load measurements to find the systematic error, calibration 509 measurements should be performed with the same measurement load that 510 the instruments will see in the field. 512 We wish to reiterate that this statistical treatment refers to the 513 calibration of the instrument; it is used to "calibrate the meter 514 stick" and say how well the meter stick reflects reality. 516 5.6.4 Errors in incT 518 The nominal interval between packets, incT, can vary during either 519 active or passive measurements. In passive measurement, packet 520 headers may include a timestamp applied prior to most of the 521 protocol stack, and the actual sending time may vary due to 522 processor scheduling. For example, H.323 systems are required to 523 have packets ready for the network stack within 5 ms of their ideal 524 time. There may be additional variation from the network between the 525 Src and the MP(Src). Active measurement systems may encounter 526 similar errors, but to a lesser extent. These errors must be 527 accounted for in some types of analysis. 529 5.7 Reporting 531 The calibration and context in which the method is used MUST be 532 carefully considered, and SHOULD always be reported along with 533 metric results. We next present five items to consider: the Type-P 534 of test packets, the threshold of delay equivalent to loss, error 535 calibration, the path traversed by the test packets, and background 536 conditions at Src, Dst, and the intervening networks during a 537 sample. This list is not exhaustive; any additional information that 538 could be useful in interpreting applications of the metrics should 539 also be reported. 541 5.7.1. Type-P 543 As noted in the Framework document [3], the value of a metric may 544 depend on the type of IP packets used to make the measurement, or 545 "type-P". The value of Type-P-One-way-Periodic-Delay could change 546 if the protocol (UDP or TCP), port number, size, or arrangement for 547 special treatment (e.g., IP precedence or RSVP) changes. The exact 548 Type-P used to make the measurements MUST be reported. 550 5.7.2. Threshold for delay equivalent to loss 552 In addition, the threshold for delay equivalent to loss (or 553 methodology to determine this threshold) MUST be reported. 555 5.7.3. Calibration results 557 + If the systematic error can be determined, it SHOULD be removed 558 from the measured values. 559 + You SHOULD also report the calibration error, e, such that the 560 true value is the reported value plus or minus e, with 95% 561 confidence (see the last section.) 562 + If possible, the conditions under which a test packet with finite 563 delay is reported as lost due to resource exhaustion on the 564 measurement instrument SHOULD be reported. 566 5.7.4. Path 567 The path traversed by the packets SHOULD be reported, if possible. 568 In general it is impractical to know the precise path a given packet 569 takes through the network. The precise path may be known for 570 certain Type-P packets on short or stable paths. If Type-P includes 571 the record route (or loose-source route) option in the IP header, 572 and the path is short enough, and all routers on the path support 573 record (or loose-source) route, then the path will be precisely 574 recorded. 576 This may be impractical because the route must be short enough, many 577 routers do not support (or are not configured for) record route, and 578 use of this feature would often artificially worsen the performance 579 observed by removing the packet from common-case processing. 580 However, partial information is still valuable context. For example, 581 if a host can choose between two links (and hence two separate 582 routes from Src to Dst), then the initial link used is valuable 583 context. {Comment: For example, with one commercial setup, a Src on 584 one NAP can reach a Dst on another NAP by either of several 585 different backbone networks.} 587 6. Additional discussion on periodic sampling 589 Fig.1 illustrates measurements on multiple protocol levels that are 590 relevant to this memo. The user's focus is on transport quality 591 evaluation from application point of view. However, to properly 592 separate the quality contribution of the operating system and codec 593 on packet voice, for example, it is beneficial to be able to measure 594 quality at IP level [6]. Link layer monitoring provides a way of 595 accounting for link layer characteristics such as bit error rates. 597 --------------- 598 | application | 599 --------------- 600 | transport | <-- 601 --------------- 602 | network | <-- 603 --------------- 604 | link | <-- 605 --------------- 606 | physical | 607 --------------- 609 Fig. 1: Different possibilities for performing measurements: a 610 protocol view. Above, "application" refers to all layers above L4 611 and is not used in the OSI sense. 613 In general, the results of measurements may be influenced by 614 individual application requirements/responses related to the 615 following issues: 617 + Lost packets: Applications may have varying tolerance to lost 618 packets. Another consideration is the distribution of lost packets 619 (i.e. random or bursty). 620 + Long delays: Many applications will consider packets delayed 621 longer than a certain value to be equivalent to lost packets 622 (i.e. real time applications). 623 + Duplicate packets: Some applications may be perturbed if 624 duplicate packets are received. 625 + Reordering: Some applications may be perturbed if packets arrive 626 out of sequence. This may be in addition to the possibility of 627 exceeding the "long" delay threshold as a result of being out of 628 sequence. 629 + Corrupt packet header: Most applications will probably treat a 630 packet with a corrupt header as equivalent to a lost packet. + 631 Corrupt packet payload: Some applications (e.g. digital voice 632 codecs) may accept corrupt packet payload. In some cases, the 633 packet payload may contain application specific forward error 634 correction (FEC) that can compensate for some level of 635 corruption. 636 + Spurious packet: Dst may receive spurious packets (i.e. packets 637 that are not sent by the Src as part of the metric). Many 638 applications may be perturbed by spurious packets. 640 Depending, e.g., on the observed protocol level, some issues listed 641 above may be indistinguishable from others by the application, it 642 may be important to preserve the distinction for the operators of 643 Src, Dst, and/or the intermediate network(s). 645 6.1 Measurement applications 647 This sampling method provides a way to perform measurements 648 irrespective of the possible QoS mechanisms utilized in the IP 649 network. As an example, for a QoS mechanism without hard guarantees, 650 measurements may be used to ascertain that the "best" class gets the 651 service that has been promised for the traffic class in question. 652 Moreover, an operator could study the quality of a cheap, low- 653 guarantee service implemented using possible slack bandwidth in 654 other classes. Such measurements could be made either in studying 655 the feasibility of a new service, or on a regular basis. 657 IP delivery service measurements have been discussed within the 658 International Telecommunications Union (ITU). A framework for IP 659 service level measurements (with references to the framework for IP 660 performance [3]) that is intended to be suitable for service 661 planning has been approved as I.380 [7]. ITU-T Recommendation I.380 662 covers abstract definitions of performance metrics. This memo 663 describes a method that is useful both for service planning and end- 664 user testing purposes, in both active and passive measurements. 666 Delay measurements can be one-way [3,4], paired one-way, or round- 667 trip [8]. Accordingly, the measurements may be performed either with 668 synchronized or unsynchronized Src/Dst host clocks. Different 669 possibilities are listed below. 671 The reference measurement setup for all measurement types is shown 672 in Fig. 2. 674 ----------------< IP >-------------------- 675 | | | | 676 ------- ------- -------- -------- 677 | Src | | MP | | MP | | Dst | 678 ------- |(Src)| |(Dst) | -------- 679 ------- -------- 681 Fig. 2: Example measurement setup. 683 An example of the use of the method is a setup with a source host 684 (Src), a destination host (Dst), and corresponding measurement 685 points (MP(Src) and MP(Dst)) as shown in Figure 2. Separate 686 equipment for measurement points may be used if having Src and/or 687 Dst conduct the measurement may significantly affect the delay 688 performance to be measured. MP(Src)should be placed/measured close 689 to the egress point of packets from Src. MP(Dst) should be 690 placed/measure close to the ingress point of packets for Dst. 691 "Close" is defined as a distance sufficiently small so that 692 application-level performance characteristics measured (such as 693 delay) can be expected to follow the corresponding performance 694 characteristic between Src and Dst to an adequate accuracy. Basic 695 principle here is that measurement results between MP(Src) and 696 MP(Dst) should be the same as for a measurement between Src and Dst, 697 within the general error margin target of the measurement (e.g., < 1 698 ms; number of lost packets is the same). If this is not possible, 699 the difference between MP-MP measurement and Src-Dst measurement 700 should preferably be systematic. 702 The test setup just described fulfills two important criteria: 1) 703 Test is made with realistic stream metrics, emulating - for example 704 - a full-duplex Voice over IP (VoIP) call. 2) Either one-way or 705 round-trip characteristics may be obtained. 707 It is also possible to have intermediate measurement points between 708 MP(Src) and MP(Dst), but that is beyond the scope of this document. 710 6.1.1 One way measurement 712 In the interests of specifying metrics that are as generally usable 713 as possible, application-level measurements based on one-way delays 714 are used in the example metrics. The implication of application- 715 level measurement for bi-directional applications such as 716 interactive multimedia conferencing is discussed below. 718 Performing a single one-way measurement only yields information on 719 network behavior in one direction. Moreover, the stream at the 720 network transport level does not emulate accurately a full-duplex 721 multimedia connection. 723 6.1.2 Paired one way measurement 725 Paired one way delay refers to two multimedia streams: Src to Dst 726 and Dst to Src for the same Src and Dst. By way of example, for some 727 applications, the delay performance of each one way path is more 728 important than the round trip delay. This is the case for delay- 729 limited signals such as VoIP. Possible reasons for the difference 730 between one-way delays is different routing of streams from Src to 731 Dst vs. Dst to Src. 733 For example, a paired one way measurement may show that Src to Dst 734 has an average delay of 30ms while Dst to Src has an average delay 735 of 120ms. To a round trip delay measurement, this example would look 736 like an average of 150ms delay. Without the knowledge of the 737 asymmetry, we might miss a problem that the application at either 738 end may have with delays averaging more than 100ms. 740 Moreover, paired one way delay measurement emulates a full-duplex 741 VoIP call more accurately than a single one-way measurement only. 743 6.1.3 Round trip measurement 745 From the point of view of periodic multimedia streams, round-trip 746 measurements have two advantages: they avoid the need of host clock 747 synchronization and they allow for a simulation of full-duplex 748 communication. The former aspect means that a measurement is easily 749 performed, since no special equipment or NTP setup is needed. The 750 latter property means that measurement streams are transmitted in 751 both directions. Thus, the measurement provides information on 752 quality of service as experienced by two-way applications. 754 The downsides of round-trip measurement are the need for more 755 bandwidth than an one-way test and more complex accounting of packet 756 loss. Moreover, the stream that is returning towards the original 757 sender may be more bursty than the one on the first "leg" of the 758 round-trip journey. The last issue, however, means in practice that 759 returning stream may experience worse QoS than the out-going one, 760 and the performance estimates thus obtained are pessimistic ones. 761 The possibility of asymmetric routing and queuing must be taken into 762 account during analysis of the results. 764 Note that with suitable arrangements, round-trip measurements may be 765 performed using paired one way measurements. 767 6.2 Statistics calculable from one sample 768 Some statistics may be particularly relevant to applications 769 simulated by periodic streams, such as the range of delay values 770 recorded during the sample. 772 For example, a sample metric generates 100 packets at MP(Src) with 773 the following measurements at MP(Dst): 775 + 80 packets received with delay [i] <= 20 ms 776 + 8 packets received with delay [i] > 20 ms 777 + 5 packets received with corrupt packet headers 778 + 4 packets from MP(Src) with no matching packet recorded 779 at MP(Dst) (effectively lost) 780 + 3 packets received with corrupt packet payload and delay [i] <= 781 20 ms 782 + 2 packets that duplicate one of the 80 packets received 783 correctly as indicated in the first item 785 For this example, packets are considered acceptable if they are 786 received with less than or equal to 20ms delays and without corrupt 787 packet headers or packet payload. In this case, the percentage of 788 acceptable packets is 80/100 = 80%. 790 For a different application which will accept packets with corrupt 791 packet payload and no delay bound (so long as the packet is 792 received), the percentage of acceptable packets is (80+8+3)/100 = 793 91%. 795 6.3 Statistics calculable from multiple samples 797 There may be value in running multiple tests using this method to 798 collect a "sample of samples". For example, it may be more 799 appropriate to simulate 1,000 two-minute VoIP calls rather than a 800 single 2,000 minute call. When considering collection of multiple 801 samples, issues like the interval between samples (e.g. minutes, 802 hours), composition of samples (e.g. equal Tf-T0 duration, different 803 packet sizes), and network considerations (e.g. run different 804 samples over different intervening link-host combinations) should be 805 taken into account. For items like the interval between samples, 806 the usage pattern for the application of interest should be 807 considered. 809 When computing statistics for multiple samples, more general 810 statistics (e.g. median, percentile, etc.) may have relevance with a 811 larger number of packets. 813 6.4 Background conditions 815 In many cases, the results may be influenced by conditions at Src, 816 Dst, and/or any intervening networks. Factors that may affect the 817 results include: traffic levels and/or bursts during the sample, 818 link and/or host failures, etc. Information about the background 819 conditions may only be available by external means (e.g. phone 820 calls, television) and may only become available days after samples 821 are taken. 823 6.5 Considerations related to delay 825 For interactive multimedia sessions, end-to-end delay is an 826 important factor. Too large a delay reduces the quality of the 827 multimedia session as perceived by the participants. One approach 828 for managing end-to-end delays on an Internet path involving 829 heterogeneous link layer technologies is to use per-domain delay 830 quotas (e.g. 50 ms for a particular IP domain). However, this scheme 831 has clear inefficiencies, and can over-constrain the problem of 832 achieving some end-to-end delay objective. A more flexible 833 implementation ought to address issues like possibility of 834 asymmetric delays on paths, and sensitivity of an application to 835 delay variations in a given domain. There are several alternatives 836 as to the delay statistic one ought to use in managing end-to-end 837 QoS. This question, although very interesting, is not within the 838 scope of this memo and is not discussed further here. 840 7. Security Considerations 842 7.1 Denial of Service Attacks 844 This method generates a periodic stream of packets from one host 845 (Src) to another host (Dst) through intervening networks. This 846 method could be abused for denial of service attacks directed at Dst 847 and/or the intervening network(s). 849 Administrators of Src, Dst, and the intervening network(s) should 850 establish bilateral or multi-lateral agreements regarding the 851 timing, size, and frequency of collection of sample metrics. Use of 852 this method in excess of the terms agreed between the participants 853 may be cause for immediate rejection or discard of packets or other 854 escalation procedures defined between the affected parties. 856 7.2 User data confidentiality 858 Active use of this method generates packets for a sample, rather 859 than taking samples based on user data, and does not threaten user 860 data confidentiality. Passive measurement must restrict attention to 861 the headers of interest. Since user payloads may be temporarily 862 stored for length analysis, suitable precautions MUST be taken to 863 keep this information safe and confidential. 865 7.3 Interference with the metric 867 It may be possible to identify that a certain packet or stream of 868 packets is part of a sample. With that knowledge at Dst and/or the 869 intervening networks, it is possible to change the processing of the 870 packets (e.g. increasing or decreasing delay) that may distort the 871 measured performance. It may also be possible to generate 872 additional packets that appear to be part of the sample metric. 873 These additional packets are likely to perturb the results of the 874 sample measurement. 876 To discourage the kind of interference mentioned above, packet 877 interference checks, such as cryptographic hash, may be used. 879 8. IANA Considerations 881 Since this method and metric do not define a protocol or well-known 882 values, there are no IANA considerations in this memo. 884 9. References 886 1 Bradner, S., "The Internet Standards Process -- Revision 3", BCP 887 9, RFC 2026, October 1996. 889 2 Bradner, S., "Key words for use in RFCs to Indicate Requirement 890 Levels", RFC 2119, March 1997. 892 3 Paxson, V., Almes, G., Mahdavi, J., and Mathis, M., "Framework 893 for IP Performance Metrics", RFC 2330, May 1998. 895 4 Almes, G., Kalidindi, S., and Zekauskas, M., "A one-way delay 896 metric for IPPM", RFC 2679, September 1999. 898 5 Demichelis, C., and Chimento, P., "IP Packet Delay Variation 899 Metric for IPPM", work in progress. 901 6 "End-to-end Quality of Service in TIPHON systems; Part 5: Quality 902 of Service (QoS) measurement methodologies", ETSI standard TS 101 903 329-5 V1.1.2 (2002-01). 905 7 International Telecommunications Union, "Internet protocol data 906 communication service _ IP packet transfer and availability 907 performance parameters", Telecommunications Sector Recommendation 908 I.380 (to be re-designated Y.1540), February 1999. 910 8 Almes, G., Kalidindi, S., and Zekauskas, M., "A round-trip delay 911 metric for IPPM", IETF RFC 2681. 913 10. Acknowledgments 915 The authors wish to thank the chairs of the IPPM WG (Matt Zekauskas 916 and Merike Kaeo) for comments that have made the present draft 917 clearer and more focused. Howard Stanislevic and Will Leland have 918 also presented useful comments and questions. We also acknowledge 919 Henk Uijterwaal's continued challenge to develop the motivation for 920 this method. The authors have built on the substantial foundation 921 laid by the authors of the framework for IP performance [3]. 923 11. Author's Addresses 925 Vilho Raisanen 926 Nokia Networks 927 P.O. Box 300 928 FIN-00045 Nokia Group 929 Finland 930 Phone +358 7180 8000 Fax. +358 9 4376 6852 931 933 Glenn Grotefeld 934 Motorola, Inc. 935 1501 W. Shure Drive, MS 2F1 936 Arlington Heights, IL 60004 USA 937 Phone +1 847 435-0730 Fax +1 847 632-6800 938 940 Al Morton 941 AT&T Labs 942 Room D3 - 3C06 943 200 Laurel Ave. South 944 Middletown, NJ 07748 USA 945 Phone +1 732 420 1571 Fax +1 732 368 1192 946 948 Full Copyright Statement 950 "Copyright (C) The Internet Society (date). All Rights Reserved. 951 This document and translations of it may be copied and furnished to 952 others, and derivative works that comment on or otherwise explain it 953 or assist in its implmentation may be prepared, copied, published 954 and distributed, in whole or in part, without restriction of any 955 kind, provided that the above copyright notice and this paragraph 956 are included on all such copies and derivative works. However, this 957 document itself may not be modified in any way, such as by removing 958 the copyright notice or references to the Internet Society or other 959 Internet organizations, except as needed for the purpose of 960 developing Internet standards in which case the procedures for 961 copyrights defined in the Internet Standards process must be 962 followed, or as required to translate it into languages other than 963 English. 965 The limited permissions granted above are perpetual and will not be 966 revoked by the Internet Society or its successors or assigns. 968 This document and the information contained herein is provided on an 969 "AS IS" basis and THE INTERNET SOCIETY AND THE INTERNET ENGINEERING 970 TASK FORCE DISCLAIMS ALL WARRANTIES, EXPRESS OR IMPLIED, INCLUDING 971 BUT NOT LIMITED TO ANY WARRANTY THAT THE USE OF THE INFORMATION 972 HEREIN WILL NOT INFRINGE ANY RIGHTS OR ANY IMPLIED WARRANTIES OF 973 MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.