Performance analysis of iterative polling scheme for real-time reliable multicast

We present a scheme for reliable multicast based on the combination of correction codes and iterative polling. The receivers are polled for the number of lost packets in a series of requests with increasing reply probability. Receivers that have a lower loss number than the ones that replied in the previous poll are excluded from the poll. This exclusion prevents an implosion of responses to the sender, while the increase in reply probabilities eventually discloses the maximum number of packets lost by a receiver in the multicast group. This information is used to determine an adequate transmission of correction code, such that the majority of receivers can repair their - possibly different losses. We analyze how the reply parameters can be chosen such that the amount of feedback satisfies given statistical bounds. The approach is based on estimating the number of receivers above a certain loss level, and does not make assumptions about the loss characteristics of the network.