A One-Way Loss Metric for IP Performance Metrics (IPPM)
Extracted elements (28)
A corrupted packet that arrives at the destination is counted as lost rather than received. If the IP header is corrupted, the source or destination address cannot be confirmed, making it impossible to associate the corrupted packet with a specific sent test packet; counting it as received would be inconsistent.
Clock synchronization errors and an insufficiently high loss threshold Tmax can cause finite-delay packets to be misclassified as lost. The loss threshold must be set high enough and clocks synchronized well enough that arriving packets are rarely counted as lost; sensitivity of loss-only measurement to clock error is less than for delay measurement.
One-way loss measurement is preferred over round-trip because Internet paths are often asymmetric, with different router sequences for forward and reverse paths, meaning round-trip measurements conflate the performance of two distinct paths. Additionally, application performance often depends on loss in a single direction.
Poisson sampling is used to schedule measurements because it avoids self-synchronization effects and produces a statistically unbiased sample. The Poisson distribution limits the bias introduced by regular periodic probing, though other methods such as periodic sampling with random start times (RFC 3432) may be appropriate for specific situations.
The loss threshold Tmax is intentionally left somewhat vague in definition, representing a value so large that any value in [Tmax-delta, Tmax+delta] is an equivalent threshold. Enforcing a specific constant waiting time on stored delay singletons is also compliant, allowing results to serve multiple reporting audiences.
The Poisson arrival rate lambda is deliberately left unconstrained except at extremes: too high a rate perturbs the network and causes congestion; too low a rate fails to capture interesting network behavior. For detecting loss ratios below 1% over one-minute periods, several hundred samples per minute may be required.
Type-P-One-way-Packet-Loss is exactly 0 when Type-P-One-way-Delay has a finite value, and exactly 1 when Type-P-One-way-Delay is undefined. This links loss measurement to the delay metric and means the Tmax threshold defines the boundary between 'very large delay' and 'loss'.
Two errata from RFC 2680 were incorporated: EID 1528 (status: Verified) and EID 3186 (status: Held for Document Update). These minor revisions were applied to Sections 1 and 2.7 of this document.
Out-of-order arrival of test packets must be handled correctly: it is possible for a packet sent second (at TS[i+1]) to arrive at the destination before a packet sent first (at TS[i]). Reordering metrics are defined in RFC 4737.
The standard-formed packet definition applicable to loss metrics (Section 15 of RFC 2330) currently applies only to IPv4; duplicated packets that arrive at the destination as multiple non-corrupt copies are counted as received.
Time-slotted links (described in RFC 7312) can greatly modify the sample characteristics of a Poisson stream: unbiased packets with randomized inter-packet intervals may be converted to a new distribution, possibly with strong periodic characteristics, after encountering a time-slotted link.
Type-P-One-way-Packet-Loss is directly related to Type-P-One-way-Delay (RFC 7679): loss is 0 exactly when delay has a finite value, and 1 exactly when delay is undefined. Practical measurement systems therefore often combine methods for measuring both delay and loss using the same Tmax parameter.
Measurement methodologies SHOULD include appropriate techniques to reduce the probability that measurement traffic can be distinguished from normal traffic, preventing routers from treating it differently and attackers from injecting false measurements.
Measurement parameters MUST be carefully selected so that the measurements inject trivial amounts of additional traffic into the networks they measure; injecting too much traffic can skew results and in extreme cases cause congestion and denial of service.
The calibration and context for the underlying singletons MUST be reported along with the Poisson stream sample, per the reporting requirements of Section 2.8.
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, instrument calibration, and path traversed.
The degree of synchronization between Src and Dst clocks MUST be reported. If possible, instances where a packet arrived at the Dst network interface but was reported as lost due to resource exhaustion SHOULD be reported.
The exact Type-P used to make the measurements MUST be accurately reported, since the metric value may depend on protocol (UDP or TCP), port number, size, DS Field, ECN, or RSVP settings.
The loss threshold Tmax (or other methodology used to distinguish finite delay from loss) MUST be reported along with the metric results.
The path traversed by the test packets SHOULD be reported if possible, including initial link selection where the host can choose between multiple links.
Active measurement techniques greatly reduce the sensitive information available to potential observers compared to passive observation of user traffic. Passive observations of user traffic for measurement purposes raise many privacy issues; implementors are referred to the LMAP Framework (RFC 7594) for coverage of both active and passive technique privacy considerations.
The sample metric parameters are: Src, Dst, T0 (start time), Tf (end time), Tmax (loss threshold), and lambda (Poisson rate in reciprocal seconds). The output is a sequence of <T, L> pairs with monotonically increasing T values.
The singleton metric parameters are: Src (IP address of sending host), Dst (IP address of receiving host), T (send time), and Tmax (loss threshold waiting time). Metric units are binary: 0 (received) or 1 (lost).
Type-P-One-way-Packet-Loss is the singleton metric defined as a binary value: 0 if Src sent a Type-P packet to Dst at wire time T and Dst received it, or 1 if Dst did not receive it within the loss threshold Tmax.
Type-P-One-way-Packet-Loss-Poisson-Stream is a sample metric consisting of a sequence of <time, loss> pairs generated by scheduling singleton measurements at times drawn from a pseudorandom Poisson process with rate lambda between T0 and Tf.
Type-P-One-way-Packet-Loss-Ratio is the average of all L values in a Type-P-One-way-Packet-Loss-Poisson-Stream, representing the ratio of lost packets to total packets. It is undefined if the sample is empty.
Collecting measurements or using measurement results for reconnaissance to assist subsequent system attacks is common; access to measurement results and control of measurement systems must be guarded against. RFC 7594 (LMAP Framework) Section 7 provides system requirements to avoid measurement system compromise.
There are two types of security concerns: harm caused by measurements (active probing injecting traffic that can cause congestion) and harm to measurements (routers giving measurement traffic special priority, or attackers injecting artificial traffic to artificially lower the measured loss ratio). Authentication techniques such as digital signatures may guard against injected traffic attacks.