The file allocation problem--A queueing network optimization approach

Abstract

A set of customers use a connected network of computer installations, each accessing the network from a particular node. These customers share information contained in a set of data files. A typical customer's need is characterized by a request requiring a subset of these files being accessed in a Markovian sequence. The cycle time for this customer is the total time taken on the average to complete his request sequence. The objective is to locate a single copy of each of the files in such a way that a weighted sum of these response times is minimized. The problem is modelled as a Closed Queueing Network optimization problem. Models are developed for both single and multiple chain cases. An incremental analysis approach is used to solve the single chain case. For the multiple chain case, it is shown how this model approximates to a set partitioning problem under certain conditions. Efficient heuristics are developed to solve this partitioning problem. With certain simplifying assumptions, the associated communication problem is then included in the model.