A 'greedy' approach to the write problem in shadowed disk systems

There has been considerable interest in parallel disk drive systems, in which replication of the stored data is used for both fault tolerance and enhanced performance. It has been discovered that performance gains are limited by the fact that no matter how large the number of disk drives, the system still behaves probabilistically like a small system. These findings are summarized, and a greedy, that is, anticipatory, approach is proposed to deal with the problem. In this approach idle read/write heads are moved to positions which will minimize the expected seek time of the next disk access request. Substantial performance gains are found to accrue.<
>