Handling large real-time disk access requests with variable priorities

Abstract

This paper addresses the problem of providing different levels of performance guarantee for disk I/O. In typical applications, disk requests are classified into different categories based on the required quality of service (QoS), which is usually characterized by a priority and a deadline for each request. Traditional algorithms usually service all high priority requests before low priority requests, and this may result in potential starvation for the low priority requests. In this paper, a disk-scheduling algorithm is introduced to provide such QoS guarantee and avoid starvation. Our target applications for this algorithm are non-linear editing systems for continuous data, where the block size is large enough to ignore the seek time. The proposed algorithm tries to service a request with lower priority and strict deadline only if servicing this request will not violate the deadline constraints of a higher priority request. Simulation experiments are presented to show the superiority of the proposed algorithm over the traditional ones.