IP Performance Metrics (IPPM): Spatial and Multicast
updated by
- rfc-6248 — RFC 4148 and the IP Performance Metrics (IPPM) Registry of Metrics Are Obsolete
Extracted elements (29)
Method 1 (each point of interest computes statistics over time, then the reference point aggregates over space) is preferred over Method 2 (raw singletons sent to reference point first) because Method 1 reporting volume depends only on computation period, not sample count.
PDV (Type-P-Segment-One-way-ipdv-min-Stream) is less sensitive to inter-packet interval variations than IPDV metrics because it reflects a wide range of spacings in any single distribution, making segment PDV results comparable to source-is-Ha measurements.
The segment IPDV inter-packet interval at Ha is not under the control of Ha because delay variation between Src and Ha will perturb the spacing; the ingress inter-packet interval must be known at Ha to fully interpret segment IPDV results.
Segment metrics SHALL be declared invalid when Ha or Hb disappears from the path or when their ordering changes due to routing changes; for packet loss, the metric is also invalid when Lk.a=1 but Lk.b=0 (packet unseen at Ha but seen at Hb).
Spatial delay values may appear to decrease (dTi > dTi+1), which is physically impossible, but can result from clock synchronization errors between measurement points, path reconvergence changing router order, or differences between wire-time and host-time at collection points.
Duplication of routers in the routers digest MUST be identified before computation of statistics to avoid corrupted information; path instability (e.g., micro-loops during reconvergence) can cause a test packet to be observed more than once by the same router.
Observation/measurement collection points and the destination SHOULD use consistent methods to detect packet losses; the methods and parameters must be systematically reported to permit careful comparison and avoid confounding factors.
Spatial one-way delay measurements MUST respect the methodology, clock, uncertainties, and reporting requirements of RFC 2679 sections 3.5–3.8, applied independently at each point of interest Hi.
Spatial packet loss measurements MUST respect the methodology, clock, uncertainties, and reporting requirements of RFC 2680 sections 2.5–2.8 for each point of interest Hi.
The packets of a Type-P stream used for Type-P-Spatial-One-way-ipdv-Vector MUST all be of the same length L.
A 'matrix' is multiple vectors organized over both time and space dimensions. Several vectors V1,...,Vm form a matrix containing results observed at different places and times, enabling both per-receiver and cross-receiver statistical analysis.
A 'routers digest' is the ordered list of routers on the path from source to destination that serve as spatial measurement collection points (points of interest). Spatial vector metrics are computed over this ordered sequence.
A 'vector' is a set of singletons from observations of a single source packet at different measurement hosts in the network. For N receivers, the delay vector V = {dT1, dT2,...,dTN} captures performance across the spatial dimension.
GLR (Group Loss Ratio) is the sum of all loss singletons Ln(k) over N receivers and K packets divided by K*N; GRLR (Range Loss Ratio) = max(RnLR) - min(RnLR), with both max and min reported individually for full characterization.
GRDV (Type-P-One-to-group-Range-Delay-Variation) = max(RnDV) - min(RnDV) over all N receivers, using the 1-10^(-3) quantile of one-way delay minus minimum delay as the per-receiver DV; the quantile used must be recorded with results.
One-to-group delay statistics define: RnMD (per-receiver mean delay = mean of finite singletons at receiver n), GMD (group mean = mean of all RnMD), GRMD (range = max(RnMD) - min(RnMD)), and GMMD (maximum = max(RnMD)).
RnCLR (Comparative Loss Ratio) uses the maximum number of packets received at any receiver as denominator rather than total packets sent, enabling comparison of relative loss performance across receivers in a group.
Type-P-One-to-group-Delay-Vector is the set of Type-P-One-way-Delay singletons {dT1,...,dTn} from source Src to N receivers; Gr (multicast group address) is an OPTIONAL parameter to differentiate measured traffic and support both multicast and unicast measurement.
Type-P-One-to-group-Packet-Loss-Vector is the set {T, L1,...,LN} of Boolean loss singletons from source Src to N receivers, where L=0 indicates packet arrival within Tmax and L=1 indicates loss at that receiver.
Type-P-Segment-ipdv-prev-Stream measures inter-packet delay variation between two path routers using consecutive packet pairs; each element is dTk.ab - dTk-1.ab, undefined if any of the four delay values at Ha or Hb are undefined.
Type-P-Segment-One-way-Delay-Stream is the sequence <dT1.ab,...,dTm.ab> where dTk.ab = dTk.b - dTk.a is the one-way delay between routers Ha and Hb extracted from the spatial delay matrix; it is undefined if the packet did not pass either router.
Type-P-Segment-One-way-ipdv-min-Stream measures segment packet delay variation (PDV) as dTk.ab - min(dTi.ab) for all packets k, where min is the minimum segment delay over the measurement interval; undefined if either endpoint delay is undefined.
Type-P-Segment-Packet-Loss-Stream between Ha and Hb is: Lk=0 if both Ha and Hb observed the packet, Lk=1 if Ha observed it but Hb did not, and undefined if neither observed it; the metric is invalid if the packet reached Dst but was deemed lost between Ha and Hb.
Type-P-Spatial-One-way-Delay-Vector decomposes end-to-end one-way delay into the sequence <T,dT1,dT2,...,dTn,dT> where dTi is the one-way delay from Src to router Hi (wire-time last bit), or undefined if the packet does not pass Hi within the loss threshold.
Type-P-Spatial-One-way-ipdv-Vector is the sequence <T1, T2, dT2.1-dT1.1, ..., dT2.n-dT1.n, dT2-dT1> representing per-hop IP packet delay variation between two packets P1 and P2; any element is undefined if either packet did not pass the corresponding router.
Type-P-Spatial-Packet-Loss-Vector is the sequence <T, L1, L2, ..., Ln> where Li=0 means the packet passed Hi with a finite delay and Li=1 means the packet did not pass Hi within the loss threshold.
Section 13 registers all spatial and one-to-group metrics defined in this memo in the IANA IP Performance Metrics Registry established by RFC 4148, including the seven spatial vector/segment metrics and the three one-to-group vector metrics with their associated statistics.
Active measurement packets used for spatial and one-to-group metrics are dedicated test packets; passive measurement of user traffic is explicitly out of scope. Implementations must ensure test traffic rates are chosen to not impact network performance.
Using Method 2 (space-first aggregation with raw singleton reporting) poses a Denial-of-Service risk toward the reference point: any increase in test packet rate directly and proportionally increases bandwidth consumption, making the measurement system impossible to dimension safely.