Summary

  • RFC 9946 defines UDPSTP for a Maximum IP-Layer Capacity measurement; its load adjustment is for infrequent, diagnostic, short-term use, not general congestion control.
  • A result can describe the tested method and conditions. It does not alone establish a subscription rate, persistent bottleneck, application outcome or service-level commitment.

RFC 9946 gives a UDP protocol to conduct the one-way IP capacity method specified by RFC 9097. That is a narrow and useful purpose. A sender adjusts offered load using receiver feedback so that the test can investigate the Maximum IP-Layer Capacity Metric. The protocol makes a measurement operation more comparable; it does not make the measured number universal.

The document marks that boundary several times. Its load-rate adjustment algorithm must not be used as a general congestion-control algorithm. It is for an infrequent, diagnostic, short-term measurement. UDPSTP itself must be used only for diagnostic and operations measurements. A fixed-rate test is for operator operation and maintenance within a local domain, while a subscriber’s diagnostic request is evidence that bottleneck capacity is uncertain, not that a promised rate has been established.

The conditions surrounding a result remain part of the result. RFC 9946 limits concurrent independent tests of a path to one. It recommends discontinuing other traffic sharing a subscriber’s dedicated resources, because the measurement may be inaccurate and elastic traffic may lose throughput. Endpoint roles, direction, server selection, offered load, feedback, authentication mode, time and competing traffic can all change what a later reader is entitled to infer.

Authentication and integrity protection for measurement feedback are recommended. They can strengthen evidence that identified test participants exchanged particular messages. They do not demonstrate a retail contract, the network outside the observed path, every bottleneck, future capacity, or an application’s performance. Those require their own evidence and their own accountable decision owner.

Al Morton edited RFC 9946 but died before it could be finalised; the RFC records that Ruediger Geib joined later to help finalise the work. Morton’s public IETF profile provides historical context, not a current role or a claim about an operator. Running-code discipline preserves the exact test receipt and its limits instead of letting a single capacity figure impersonate a service promise.

Sources