Approximation of Seek Time for Merged Disks

摘要

Seek time is among the most important performance parameters of rotational magnetic disks. The purpose of this paper to to gain further understanding of the average seek time behavior when multiple data access streams from different disks are merged. A new probabilistic model of seek time is proposed and some interesting properties are derived. Among them: arm contention is worst if the base system has uniform skew; arm contention increases after data merging; and the average seek time for the target disk is independent of the order that data access streams are merged. A counter-example is also given to show that monotone property of average seek time is not preserved. A method to determine the average seek time is presented when an arbitrary number of disks is merged or migrated.