Adaptive task farm implementation strategies

Abstract

Task farm parallelism exploitation pattern is used in a wide range of parallel applications. Due to the independence of the computations relative to the different tasks, efficient methods have been developed that can be used to implement task farms on both SMP/MPP and on homogeneous, static network/clusters of workstations. In this work we discuss a new implementation strategy for task farm computations that takes into account dynamicity and heterogeneity in the availability of computing resources as well as fault tolerance. The proposed implementation schema suitably addresses classical situations relative to both peer-to-peer and GRID worlds.