Summary

  • RFC 6374 observes a pre-existing MPLS LSP, pseudowire or section. It supports proactive and on-demand operation, direct and inferred loss measurement, and in-band or out-of-band exchanges.
  • Direct loss can provide exact counter accounting but may require specialized hardware. Inferred loss uses test traffic and is approximate evidence, not an equivalent to data-plane counters.
  • One-way delay needs synchronized clocks. Two-way metrics avoid direct comparison of non-local clocks; two-way channel delay is not automatically the same as round-trip delay.

The querier selects the measurement type, scope and rate. The responder supplies counters or timestamps. Implementations determine supported timestamp formats, measurement points and scopes and must disclose behavior clearly. IETF and IANA define interoperable encodings, including distinct G-ACh Channel Types for direct loss, inferred loss, delay and combined measurements; registration does not prove receiver support or enablement. Higher-level management sets thresholds, interprets SLAs and decides whether to act.

Meaning is conditional. A counter may describe a particular traffic class, ingress-to-egress span or maintenance segment rather than every packet using the service. Measurement-point placement, ECMP, counter alignment, loss scope, timestamp format, synchronization and clock quality can all change the conclusion. A single sampled ECMP path must not be presented as every alternate path. The standards do not quantify deployment accuracy without those details.

Proactive measurement consumes traffic and processing. Rates should fit the context. A querier should suspend sustained measurement when lost measurement messages exceed its threshold, and a receiver should rate-limit excessive requests. RFC 6375 profiles the method for MPLS-TP: configuration is external, defaults are fixed so negotiation is unnecessary, received values should be validated, and the operator controls rates. RFC 7876 adds a UDP Return Object for an explicitly selected out-of-band response path; it changes return transport, not measurement semantics, and requires reflection-abuse prevention.

Protected out-of-band returns, such as suitable IPsec protection where needed, may be necessary.

This is not authentication or authorization. LM/DM assumes path integrity and far-end identity were independently verified by existing mechanisms. Implementations must support disabling each Channel Type when that assurance is absent. Even with identity verification, injection or man-in-the-middle interference can corrupt evidence. A valid response therefore does not establish an LSP, reroute it, reserve bandwidth, repair it, change policy or authorize remediation.

Verification fixture: define the exact LSP, pseudowire or section; record traffic class, ECMP treatment and measurement-point locations; capture aligned counter snapshots; test each supported Channel Type and timestamp format; compare one-way results only after verifying clock synchronization; calculate two-way channel and round-trip values separately; inspect lost-message suspension and receiver rate limiting; validate external RFC 6375 configuration; and test that an RFC 7876 UDP return follows the requested path without becoming an amplifier. Record implementation defaults only as observed facts, never as standards requirements.

Operator decision path: first verify identity and path integrity; next choose direct or inferred loss and the required scope; then select a rate that the path can carry; validate counters, clocks and return protection; establish a baseline; compare results with a separately owned threshold; and finally let the management system decide investigation, SLA handling or remediation. The protocol supplies evidence, not the decision.

Without MPLS-native LM/DM, teams would lean more on IP probes, indirect symptoms or vendor-specific mechanisms. Those alternatives may not share the measured forwarding context or support direct loss accounting. They can still be useful, but their evidence is not interchangeable.

Sources