Optimal Deterministic Timeouts for Reliable Scalable Multicast

Reliable multicast protocols suuer from the problem of feedback implosion. To avoid this problem, the number of receivers sending feedback in case of loss must be small. However, losses experienced by diierent receivers are strongly correlated, since receivers share common resources in the multicast tree. One approach to feedback implosion avoidance relies on delaying feedback at the receivers. We present DTRM (De-terministic Timeouts for Reliable Multicast), a distributed algorithm to compute optimal deterministic timeouts for each receiver in a multicast tree as a function of the tree topology and the sender-to-receiver round-trip delays. DTRM has several desirable properties. First, feedback im-plosion is provably avoided for a single loss anywhere in the tree, provided delay jitter is bounded. Second, the computation of the timeouts can be entirely distributed; receivers and intermediate nodes only rely on local topology information. Third, the timeouts computed by DTRM are optimal with respect to the maximum response time.