IP Packet Delay Variation Metric for IP Performance Metrics (IPPM)
Extracted elements (21)
Lost packets cause ipdv to be undefined, which prevents computing simple statistics such as the sample mean. The recommended approach is to condition the statistics on packet pairs where both packets arrive, effectively reducing the sample space; this yields valid conditional statistics while acknowledging the limitation.
Poisson sampling is required for the sample metric to ensure statistically unbiased, uniformly distributed sampling of transmission times across the interval [T0, Tf] and to avoid self-synchronization artifacts. Periodic (constant-bit-rate) sampling is noted as an alternative but risks aliasing.
The document avoids the term 'jitter' because it is ambiguous: it means either variation relative to a clock signal (used in synchronous/circuit contexts) or variation relative to a reference metric such as average or minimum delay (used by computer scientists). 'Delay variation' is used instead for precision.
The ipdv metric is defined as a difference of one-way-delays rather than an absolute delay measure, making it robust to clock synchronization errors. Constant clock offsets cancel out in the subtraction; only relative skew and drift during the measurement interval introduce residual error.
The metric is deliberately parameterized rather than fixed because uses of ipdv remain partly a research topic and because different communities (e.g., ITU-T Y.1541 and Demichelis) propose different definitions. Parameterization allows a single framework to encompass competing proposals as long as all parameters are reported.
When clocks are unsynchronized and skew is linear, the skew parameter S can be estimated from the sample-mean inter-packet time Ti and the sample-mean ipdv value; individual ipdv values can then be corrected by subtracting S * ti. Skew displacement does not change the shape of the distribution (e.g., standard deviation is unaffected); only drift distorts it.
The sample metric definition encompasses both the ipdv definition proposed in ITU-T Draft Y.1541 and the instantaneous ipdv definition by Demichelis (World Telecommunications Congress 2000), by choosing the appropriate selection function F.
All packets in the Type-P packet stream from which the sample ipdv metric (Type-P-One-way-ipdv-Poisson-stream) is taken MUST all be of the same length.
Any report of the ipdv metric MUST include all parameters associated with it so that the conditions and meaning of the metric can be determined exactly.
ipdv measurement packets carry no user payload information, so privacy of user data is not a concern.
One-way-ipdv values are bounded by the interval (L-U, U-L), where U is the upper bound on one-way-delay (the loss timeout) and L is the lower bound (propagation + transmission + nodal transit delay with no queuing). ipdv values within any finite interval can be positive, negative, or zero.
The selection function F specifies which two packets from the stream are used for a singleton ipdv measurement. Examples include consecutive Type-P packets, packets with specified indices, or packets with the minimum and maximum one-way-delays within the interval.
The Type-P-One-way-ipdv metric is parameterized by: Src and Dst IP addresses, times T1 and T2, packet length L in bits, a selection function F that unambiguously identifies the two packets from the stream, interval endpoints I1 and I2, and packet type specification P.
Type-P-One-way-ipdv-inverse-percentile: given a value Y, the percentage of ipdv sample values less than or equal to Y.
Type-P-One-way-ipdv is defined as the difference between the one-way-delay of two selected packets from Src to Dst: specifically, dT2 minus dT1, where dT1 and dT2 are the one-way-delays of the first and second selected packets respectively. The value is a real number of seconds (positive, zero, or negative), or undefined if one or both packets are not received.
Type-P-One-way-ipdv-jitter: the selection function specifies consecutive packet pairs; the absolute value of each ipdv is taken. An alternate formulation from RFC 1889 applies an exponential filter (j_new = 15/16 * j_old + 1/16 * |ipdv_new|) to this sequence.
Type-P-One-way-ipdv-percentile: the Xth percentile (0%–100%) of all ipdv values in a given Type-P One-Way-ipdv sample; the 50th percentile is the median.
Type-P-One-way-ipdv-Poisson-stream is a sample metric defined as a sequence of triples (T1, T2, dT), where T1 and T2 are transmission times of the first and second packets of each singleton pair and dT is the ipdv value in seconds or undefined. The stream is generated by a pseudo-random Poisson process with average rate lambda over [T0, Tf].
Type-P-One-way-peak-to-peak-ipdv: the selection function selects the packet with maximum one-way-delay as the first of each pair and the packet with minimum one-way-delay as the second, within each sub-interval. The resulting sequence captures the peak-to-peak delay variation in each sub-interval.
Adversaries may attempt to disrupt ipdv measurements by diverting or corrupting test packets. Packet authentication and integrity checks may be used to verify that test packets have not been altered in transit.
Sending large numbers of measurement packets could constitute a denial-of-service attack on the network. Measurement parameters must be carefully selected to avoid interfering with normal traffic.