Task allocation in distributed multimedia systems based on the host-satellite model

Abstract

Multimedia applications require intermediate processing between media sources and sinks. In addition to end-user machines intermediate computers can be used for performing media processing. This possibility leads to the problem of allocating processing components on various computers. In this paper, we study this problem in the context of star-shaped application graphs which have to be allocated between given end-user machines (satellites) and a central computer (host). The problem is formulated in terms of best achievable bottleneck resource usage. Several approaches are considered including anapproximate scheme and two fast-heuristics. Performance measurements show the efficiency of the considered approaches. A discussion of our approach shows important differences to solutions provided for related problems of graph partitioning and mapping.