SPC2-VOD workload

Figure 13: We measure the achieved throughput per stream as the number of concurrent video-on-demand streams increases using five SCSI disks and a cache of $100$ MB. The SPC2-VOD workload uses read size of $256$ KB, thinktime = $333.3$ ms.

In Figure 13 we measure the performance for video-on-demand workloads. The goal is to provide each sequential stream its requisite bandwidth ($768$ KBps in the case of SPC2-VOD workload) for the maximum number of streams. We can easily observe that AMP is able to entertain the most number of concurrent streams (up to $125$) at the desired bandwidth. FA $_\textrm{64/31}$ starts failing at about $100$ streams and FA $_\textrm{256/127}$ fails after $75$ streams because of more severe prefetch wastage. None of the other algorithms can match the demanded rate as they incur expensive read misses which stall the client and lower the throughput.