Disk scheduling with future knowledge

Disk scheduling is a major concern in heavily loaded systems since disk I/O is typically a performance bot­ tleneck. There have been numerous studies of the rela­ tive performance of the traditional disk scheduling al­ gorithms, SSTF, SCAN, and FCFS, as well as several studies on the more recently developed VSCAN. We present an approach to studying disk scheduling that compares each algorithm’s performance to an optimal, perhaps unattainable, performance. We consider op­ timal to mean the order of service which yields the smallest average seek distance, and we show seek dis­ tance to be a reliable measure of performance. This approach is in the same spirit as Belady’s MIN algo­ rithm for optimal page replacement in a virtual mem­ ory system. Using a modified version of a disk sim­ ulator from a previous study, we generate fixed but realistic request streams and show how each test case can be represented as a search tree. A branch and bound tree searching algorithm is used to select the optimal path. We then compare the performance of traditional disk scheduling algorithms to that of the optimal schedule. Our results indicate that there is substantial performance gain to be achieved by devel­ oping belter disk scheduling algorithms.