Towards efficient and practical reliable bulk data transport for large receiver sets.

Many critical network applications require the transmission of bulk data to a large, heterogeneous, asynchronous receiver set. Standard unicast solutions exhibit poor scaling due to inefficient use of bandwidth over shared links, prompting consideration of multicast and peer-to-peer systems. Unfortunately, these approaches introduce their own problems. In multicast, we must provide transport layer services, such as reliability and congestion/flow control. To deal with these, researchers have proposed the use of several layered multicast scheduling techniques using cyclic transmission and FEC. For peer-to-peer systems, we must address the problem of block location and extinction. Work in network coding provides an elegant solution to these problems; however, a naive implementation of such coding is computationally expensive. We propose a practical implementation of network coding. Next we compare several layered encoding schemes. Finally, we compare the performance of layered multicast and network coding in peer-to-peer systems.