Estimating the network delays between each pair of nodes in a multicast session is the key parameter in reliable multicast; it is used, among other things, in suppressing the implosion of request and repair packets, and in detecting congestion. Existing implementations use O(n) multicasts with O(n) message size each (total of O(n2) bits); here, n is the session size. We present a new protocol that requires O(n) multicasts only with O(1) message size each. We also present another protocol that estimates delays from each receiver to the sender without causing feedback implosion. Because of reduced message complexity and feedback implosion, these protocols enhance the scalability of reliable multicast. Furthermore, they do not require synchronized clocks, or any knowledge of network topology or the session size.
Volkan Ozdemir, S. Muthukrishnan, Injong Rhee