2. Approach

QProt separates mechanism from policy. Mechanism identifies flows, maintains per-flow state, and accumulates per-flow queuing scores. Policy decides when action is required and what action to take.

Per-flow state is normally keyed by an end-to-end Layer 4 5-tuple. The state contains a queuing score that decays over time. The score is transformed into time units so it can be represented as the flow state's expiry time, t_exp. Non-queue-building flows tend to release state quickly; queue-building flows hold state persistently.

The queuing score accumulates packet size scaled by probNative, the same internal variable the LL AQM uses for ECN marking. Therefore a flow's score grows faster when queuing is higher and when the flow sends faster during queuing. The score is aged at a constant AGING rate. The default described in the RFC is 2^19 B/s, about 4 Mb/s.

The DOCSIS policy action is reclassification into the Classic queue.