Vector assignment problems: A general framework

Abstract

We present a general framework for vector assignment problems. In such problems one aims at assigning n input vectors to m machines such that the value of a given target function is minimized. While previous approaches concentrated on simple target functions such as max-max, the general approach presented here enables us to design a PTAS for a wide class of target functions. In particular we are able to deal with non-monotone target functions and asymmetric settings where the cost functions per machine may be different for different machines. This is done by combining a graph-based technique and a new technique of preprocessing the input vectors.