A One-way Delay Metric for IPPM
obsoleted by
- rfc-7679 — A One-Way Delay Metric for IP Performance Metrics (IPPM)
Extracted elements (25)
One-way delay measurement is preferred over round-trip delay because Internet paths are often asymmetric (different routers forward vs. reverse), and even symmetric paths may have asymmetric queueing; round-trip measurements conflate the performance of two distinct paths.
Padding in test packets SHOULD be filled with randomized bits rather than fixed patterns to avoid artificially lowering the measured delay due to compression techniques along the path.
Poisson sampling is used to schedule test packets to avoid self-synchronization effects and to obtain a statistically unbiased sample. Other sampling methods may be appropriate in other situations and are encouraged to be submitted for standardization.
The 95% confidence level for calibration error was chosen because some confidence level is needed to remove outliers, a specific level must be named so independent implementations can be compared, and 95% was loose enough to exclude outliers even in prototype user-level implementations.
When a packet is duplicated along the path and multiple non-corrupt copies arrive, the packet is counted as received and the first copy determines the one-way delay; if a fragmented packet fails reassembly, it is deemed lost.
The simple 255-second theoretical upper bound on IP packet lifetimes (RFC 791) could be used as the loss threshold, but implementors should apply good engineering judgment about packet lifetimes in practice; for many applications, treating a very large delay as infinite causes no harm.
For the Poisson stream metric, implementors MUST report the calibration and context for the underlying singletons along with the stream.
Measurement methodologies SHOULD include appropriate techniques to reduce the probability that measurement traffic can be distinguished from normal traffic; authentication techniques such as digital signatures MAY be used to guard against injected traffic attacks.
Measurement parameters MUST be carefully selected so that the measurements inject trivial amounts of additional traffic into the networks they measure, to avoid skewing results or causing congestion and denial of service.
Systematic error SHOULD be removed from measured values before reporting; the calibration error e at 95% confidence SHOULD be reported; and conditions under which finite-delay packets may be reported as lost due to resource exhaustion SHOULD be reported.
The calibration and context in which the metric is measured MUST be carefully considered, and SHOULD always be reported along with metric results, including Type-P, loss threshold, calibration error, and path information.
The exact Type-P (protocol, port, size, precedence) used to make the measurements MUST be accurately reported, since the metric value may depend on packet type.
The threshold (or methodology to distinguish) between a large finite delay and loss MUST be reported along with measurement results.
Privacy concerns of network measurement are limited by the active measurement approach: unlike passive measurements, active measurements do not risk releasing existing user data.
Calibration decomposes measured values as: measured value = true value + systematic error + random error. The calibration error e is defined such that the reported value falls within (true value ± e) at least 95% of the time.
Clock uncertainty is characterized by four properties: synchronization (how closely two clocks agree), accuracy (agreement with UTC), resolution (clock precision), and skew (rate of change of accuracy or synchronization over time). All four contribute to measurement error.
The total clock-related measurement uncertainty is bounded by Esynch(t) + Rsource + Rdest, where Esynch(t) is an upper bound on synchronization uncertainty and Rsource, Rdest are the clock resolutions of source and destination hosts.
Type-P-One-way-Delay-Inverse-Percentile is the fraction of dT values in the stream that are less than or equal to a given time-duration threshold, ranging from 0% to 100%. It is undefined if the sample is empty.
Type-P-One-way-Delay is the singleton metric: for a real number dT, it means Src sent the first bit of a Type-P packet at wire-time T and Dst received the last bit at wire-time T+dT. The value is undefined (informally infinite) if the packet is not received.
Type-P-One-way-Delay-Median is the median of all dT values in the stream, treating undefined values as infinite. When the sample has an even number of values, the mean of the two central values is used.
Type-P-One-way-Delay-Minimum is the minimum of all dT values in the stream, treating undefined values as infinite. It is undefined if the sample is empty or all values are undefined.
Type-P-One-way-Delay-Percentile is defined as the Xth percentile of all dT values in a Poisson stream, treating undefined (lost) values as infinitely large. The statistic is itself undefined if the sample is empty.
Type-P-One-way-Delay-Poisson-Stream is the sample metric: given Src, Dst, Type-P, start time T0, end time Tf, and rate lambda, it is a sequence of <time, delay> pairs derived from a pseudo-random Poisson process of rate lambda between T0 and Tf.
Wire-time vs. host-time uncertainty is characterized by Hsource and Hdest (upper bounds on the difference between wire time and host time at source and destination), contributing a total uncertainty of Hsource + Hdest to the measurement.
Active measurements inject packets into the network and can harm it if measurement rates are too high, causing congestion or denial of service; conversely, routers may give measurement traffic different priority or an attacker may inject fake measurement traffic to skew results.