Round-Trip Packet Loss Metrics
Extracted elements (26)
Round-trip loss is asserted to be symmetric: Type-P-Round-trip-Loss(Src->Dst->Src) = Type-P-Round-trip-Loss(Dst->Src->Dst). This simplification (following RFC 2681) introduces minor ambiguity but is justified by measurement efficiency.
Round-trip metrics are motivated by the near-universal need for two-way communication (e.g., TCP three-way handshake) and the practical simplicity of echo-based test systems; they provide a basis for inferring application performance more easily than one-way measurements.
Separate loss reporting is required because many measurement systems report statistics on a conditional delay distribution (conditioned on packet arrival), which excludes lost packets; lost packets must therefore be accounted for by a standardized loss metric.
Reordering on the Src->Dst path causes misalignment of sequence numbers at Dst relative to the sender, complicating one-way inference from round-trip measurements; RFC 4737 reordering metrics and their implications were not available when RFC 2680 and RFC 2681 were published.
The reverse-path (Dst->Src) sampling process in a round-trip measurement is conditional on successful delivery to Dst; when appreciable forward-path loss or delay variation occurs, the reverse-path sample is significantly distorted, making one-way reverse-path performance assessment invalid.
TWAMP (RFC 5357) is cited as the reference measurement protocol for collecting all four timestamps needed to describe round-trip delay and remove processing time at the responding host; ICMP may be used if the issues in RFC 2681 Section 2.6 are alleviated.
A responder serving typical round-trip packet loss testing SHOULD produce a response in 1 second or less; a responder unable to satisfy this requirement SHOULD log the fact so an operator can adjust load and priorities.
Analysis of responder timestamps that finds responses are not generated in a timely fashion SHOULD result in operator notification, and the operator SHOULD suspend tests to the responder since it may be overloaded.
Every received test packet MUST result in a responding packet, and the response MUST be generated as quickly as possible, implying interface buffers are serviced promptly and buffer discards are extremely rare.
In passive measurement scenarios, since user payloads may be temporarily stored for length analysis, suitable precautions MUST be taken to keep payload information safe and confidential; a hashing function is recommended for payload comparisons.
Measurement equipment MUST be calibrated according to RFC 2679 Section 3.7.3 under a representative measurement load; both unexpected packet discards and systematic/random errors MUST be recorded.
Measurement implementations MUST address the possibility of packet reordering and avoid related errors (e.g., prematurely declaring loss when a sequence gap is observed) in their processes.
Prior to conducting measurement, the participating hosts MUST be configured to send and receive test packets of the chosen Type-P.
T0, the start time for Periodic Streams, MUST be selected at random from the interval [T, T+dT].
When a non-standard sample process is used, its details MUST be reported with the measurement results.
When deriving one-way metrics from round-trip measurements, implementations MUST address the possible outcome of packet reordering causing misalignment of sequence numbers at the Dst.
Common metric parameters include Src and Dst (IP addresses), T and Tf (test interval bounds), Tmax (maximum waiting time to distinguish loss from extreme delay), M (packets sent), N (packets received), TstampSrc and TstampDst (wire-times at each measurement point), and Type-P (packet type affecting network treatment).
Each Type-P-Round-trip-Loss singleton is represented as a pair (TstampSrc, L), where TstampSrc is the wire-time the packet left Src and L is 0 (success) or 1 (loss).
Metric units for a single packet's loss performance are logical values: 0 for successful transmission and 1 for packet loss. Time units follow RFC 2330 conventions.
Type-P-Round-trip-Loss is the singleton metric: a binary value (0 or 1) where 0 indicates the packet completed the round-trip from Src to Dst and back before TstampSrc + Tmax, and 1 indicates the reflected packet was not received within that window.
Type-P-Round-trip-Loss-<Sample>-Ratio is the primary statistic: the average of all L values in the stream, expressed in units of lost packets per round-trip transmissions attempted. It is undefined if the sample is empty.
Type-P-Round-trip-Loss-<Sample>-Stream is a sequence of (TstampSrc, L) pairs obtained from a Poisson or Periodic sampling process; the variable <Sample> SHALL be replaced with the process name used.
No IANA action is requested. The IANA IPPM Metrics Registry previously used for IETF-defined metrics was declared obsolete in RFC 6248; the metrics in this document may be considered for future registration but are not registered at this time.
Active measurement does not threaten user data confidentiality since it generates synthetic test packets rather than sampling user traffic. Passive measurement must restrict attention to headers of interest and protect any temporarily stored payload data.
An adversary who identifies test packets can manipulate their processing (increasing or decreasing delay) to distort measurements, or inject additional packets that appear to be part of the sample. Authentication or encryption (e.g., digital signatures as in RFC 5357) MAY be used to guard against injected traffic attacks.
The measurement method — sending packet streams between hosts through intervening networks — could be abused for denial-of-service attacks against the destination or intervening networks. Bilateral or multilateral agreements on timing, size, and frequency of measurements are recommended.