Optimal file allocation in a distributed computer network by orthogonal array experiments

Abstract

The problem of allocating files in a distributed network of interconnected computers, is one of minimizing the overall operating cost while satisfying the finite computer storage capacity constraint and the bounds on the time to access the files. The overall operating cost consists of the cost of storing the files in the computers and the cost of transferring them across the network. This problem has been formulated previously as a nonlinear zero-one integer programming problem (W.W.Chu, 1969) and solved as a linear zero-one integer programming problem, using the Gomory cutting technique (1963). When the communication link capacities are non-uniform, the dimension, of the problem increases drastically, and conventional integer programming techniques are not computationally efficient. The author provides an alternative approach, in which he solves the nonlinear integer problem directly, using what is known as the orthogonal array experiment. Also known as the Taguchi method used in quality control, the orthogonal array experiment is extremely efficient and, for this class of problems, provides the same optimal allocation as do compositionally intensive conventional integer programming methods.