Disk performance enhancement through Markov-based cylinder remapping

A scheme for disk subsystem performance enhancement that is based on (virtual) cylinder remapping is proposed. A natural workload on a real system is measured, and statistical tests are used to determine that disk accesses are appropriately modeled by a first order Markov chain. Maximum likelihood estimators of the Markov model parameters are used in a simulated annealing algorithm to find a permutation of the (virtual) cylinders that substantially reduces expected seek distance. This permutation is then installed in a real system and tested under a workload that is stochastically generated from the Markov model. The proposed scheme is seen to offer a 25.6% reduction in mean service time when compared to the original (unmapped) cylinder arrangement.

• Type: Conference
• Year: 1992
• Where: ACMSE
• Authors: Robert Geist, Darrell Suggs, Robert G. Reynolds, Shardul Divatia, Fred Harris, Evan Foster, Priyadarshan Kolte
• Keywords: ACMSE 1992, Markov Model, Markov Model Parameters, Order Markov Chain, Theoretical Computer Science