Record Allocation for Minimizing Expected Retrieval Costs on Drum-Like Storage Devices

Abstract

This paper examines the problem of distributing a set of equal-size records among the sectors of a drum-like storage device in order to exploit known access frequencies and reduce the average access time. A simple catenated search model is defined for which, the problem is shown to be NP-complete. Heuristics are then defined and analyzed in terms of worst-case bounds. It is shown that easily implemented highest-access-frequency-first assignment rules provide an average access time very close to optimal.