A reliability window for flexible and scalable multicast services

Abstract

This paper proposes a simple mechanism to allow reliability to be traded off against performance and scalability in a reliable multicast transport protocol. A reliability window is used in the data packet header to indicate which packets are useful to recover if lost. In addition, we describe a simple API that allows applications to set the reliability window in a way meaningful to the application. The usefulness of the trade-off is demonstrated using experimental results from a test network. We expect this mechanism to be useful for large-scale content distribution where the content ages over time.