Reliably networking a multicast repository

In this paper we consider the design of a reliable multicast facility over an unreliable multicast network. Our multicast facility has several interesting properties: it has different numbers of clients interested in each data packet, allowing us to tune our strategy for each data transmission; it has recurring data items, so that missed data items can be rescheduled for later transmission; and it allows the server to adjust the scheduler according to loss information. We exploit the properties of our system to extend traditional reliability techniques for our case, and use performance evaluation to highlight the resulting differences. We find that our reliability techniques can reduce the average client wait time by over thirty percent.