The increasing deployment of broadband services that are inherently broadcast-capable has made wide-area data broadcast an attractive data delivery alternative for large client populations. There has been significant work towards developing on-line broadcast scheduling algorithms for systems where all data items are readily available in the server's main memory. These studies ignore the data management issues that arise when the data to be broadcast must be obtained from secondary storage or remote locations. In this paper, we propose three complementary solutions to such data staging concerns: opportunistic scheduling, server caching, and prefetch-ing. These techniques exploit hints provided by the scheduling algorithm. A detailed performance evaluation using an IP Multicast-based testbed shows that these data staging techniques can dramatically enhance the performance of a large-scale on-demand broadcast system.
展开▼