Disk scheduling revisited: can O(n2) algorithms compete?

Abstract

A simple O(n2) disk scheduling algorithm is suggested, and its performance on a real system, an Intel P6-based system running Linux 2.0.18, is compared with that of standard O(n) scheduling algorithms, with the aim of determining whether the large disparity between processor speeds and disk speeds might now allow for improved disk scheduling under heavy loads. The simple O(n2) algorithm is seen to provide some modest performance improvements over the best linear algorithms. The study also identifies several areas where potentially large additional performance improvements may be found by extending the framework suggested here.