The egoistic approach to parallel process migration into heterogeneous workstation network

Abstract

A new approach to the allocation of processes in a distributed system is discussed. The proposed solution deals with process migration into heterogeneous systems by means of a strategy that delegates the individual parallel applications to manage the migration of their processes by themselves, on the basis of their own performance objectives. This approach is discussed in opposition to the global scheduling based one, and the load balancing objective is pursued as an effect of the optimization of individual applications. A new performance evaluation criterion is introduced that consists in monitoring the delays that occur when two parallel processes run towards a common synchronization point. The experiments, that have been carried out for a Simulated Annealing application, prove that a suitable setting of delay thresholds enable the application-managed migrations to lead the application and the whole distributed system to configurations of processes that are stable and effective.