Network performance measurement with periodic streams
Extracted elements (26)
Paired one-way measurement (bidirectional separate streams) is preferred over round-trip for applications like VoIP because per-direction delay asymmetry can be hidden in round-trip averages, masking application-impacting problems.
Periodic sampling is chosen over Poisson sampling for multimedia use cases because it simulates CBR traffic, is applicable to both active and passive measurements, simplifies analysis of time-varying impairments, and enables frequency-domain analysis with equally-spaced samples.
Random start times and finite-length tests are mandated to mitigate the predictability problem inherent in periodic sampling, which could otherwise allow synchronization with other periodic behavior or manipulation of results.
Separate measurement points MP(Src) and MP(Dst) are placed close to the egress/ingress of Src/Dst respectively so that the measurement apparatus itself does not significantly perturb the delay being measured.
The parameter dTloss (maximum wait time before declaring loss) is defined so implementations can treat very large delays as losses; for many real-time applications, a packet arriving after several RTT multiples is effectively as useless as a lost one.
The calibration error 'e' and systematic error bounds are specified at 95% confidence to allow comparison of results between independent implementations, mirroring the approach taken in RFC 2679.
The periodic stream method is compatible with passive measurement as well as active measurement; in passive measurement, arrival randomness of user media flows may satisfy the independent-start-time requirement, or a randomized measurement start within a flow may be needed.
For passive measurement where user payloads may be temporarily stored for length analysis, suitable precautions MUST be taken to keep this information safe and confidential.
If systematic error can be determined, it SHOULD be removed from measured values and the calibration error 'e' SHOULD also be reported such that the true value is the reported value ± e with 95% confidence.
Start times in successive time intervals MUST use an independent value drawn from the distribution, ensuring each sample interval has an unpredictable start time.
T0, the start time for generating packets, MUST be selected at random from the interval [T, T+dT]. This randomization prevents predictability and mitigates synchronization with other periodic network behavior.
The calibration and context in which the measurement method is used MUST be carefully considered, and SHOULD always be reported along with metric results.
The exact Type-P (protocol, port, size, special treatment) used to make the measurements MUST be reported, as metric values may depend on the type of IP packets used.
The threshold for delay equivalent to loss, or the methodology to determine this threshold, MUST be reported alongside measurement results.
Passive measurement may temporarily store user payload data for length analysis; suitable precautions MUST be taken to keep this information safe and confidential. Active measurement generates its own packets and does not threaten user data confidentiality.
Delay singletons cannot be computed for spurious packets (no Tstamp(Src)), unreceived or corrupt-header packets (no Tstamp(Dst)), or duplicates (only the first non-corrupt copy gets Delay[i] computed).
Global metric parameters include Src and Dst IP addresses, IPV (IP version), dTloss (maximum waiting time before declaring a packet lost), and packet size p(j) indexed by size j.
Inter-Packet Delay Variation (IPDV) singleton is defined as IPDV[i] = Delay[i] - Delay[i-1] for each packet except the first; it may be negative, zero, or positive. An example sample metric is RangeIPDV = max(IPDV[]) - min(IPDV[]).
Parameters collected at MP(Src) include Tstamp(Src)[i] per packet, with optional PktID(Src)[i] (sequence number) and PktSi(Src)[i] (actual packet size). Parameters at MP(Dst) mirror these and add optional PktStatus[i] with possible values: OK, corrupt header, corrupt payload, duplicate, fragment.
The measurement procedure collects packets from T0 to Tf at MP(Src), and from T0 to (Tf + dTstop) at MP(Dst). Data consolidation occurs at Tf + Tcons, with the constraint that dTcons >= dTstop to avoid premature collection termination.
The metric Type-P-One-way-Delay-Periodic-Stream is the primary sample metric defined, analogous to Type-P-One-way-Delay-Poisson-Stream from RFC 2679, with delay singleton Delay[i] = Tstamp(Dst)[i] - Tstamp(Src)[i].
The periodic sampling configuration is defined by five parameters: T (interval beginning), dT (duration of allowed start times), T0 (randomly selected start time from [T, T+dT]), Tf (stop time, greater than T0), and incT (nominal inter-packet interval, first bit to first bit).
Total clock-related measurement uncertainty is bounded by Esynch(t) + ResMP(Src) + ResMP(Dst) + Hsource + Hdest, combining synchronization error, clock resolution at each measurement point, and wire-vs-host time uncertainty at each end.
This document has no IANA considerations — the method and metric do not define a protocol or well-known values, so no IANA registry actions are required.
An adversary who identifies that a packet stream is part of a measurement sample may manipulate processing (e.g., artificially alter delay) or inject spurious packets to distort results. Cryptographic hash-based packet integrity checks MAY be used to discourage such interference.
The periodic stream method can be abused for denial-of-service attacks against Dst and/or intervening networks. Administrators SHOULD establish bilateral or multi-lateral agreements on timing, size, and frequency; unauthorized use may be cause for immediate packet rejection or other escalation.