Exploiting caching for realtime multimedia systems

Abstract

Several caching schemes for realtime multimedia systems have been proposed but they only concentrate on the hit ratio as the performance metric, without providing any means to utilize the disk bandwidth gained through cache hits. A high cache hit ratio alone does not capture the effectiveness of a caching scheme for such systems unless hiccup-free service is guaranteed. We evaluate three caching schemes which make it possible to provide service to additional streams with the saved disk bandwidth. We first start with NIC in which no cached intervals are preempted or replaced unless they have finished their playback. Next, we propose PSIC, which preempts the cached intervals for more valuable intervals but still guarantees hiccup-free playback. Finally, we present SPT-L scheme which has a significantly greater overhead and the best hit ratio. However STP-L suffers from frequent starvation, which verifies our claim that the high cache hit ratio does not necessary result in better performance.