Experimental design and evaluation in a distributed environment using a genetic algorithm with static allocation

We introduce in this paper an execution model for parallel applications in distributed environments. The model is then transposed into a framework that uses two main subsystems to facilitate fast applications development and complete runtime management for parallel tasks and the data flow. The framework has been evaluated by considering three groups of applications with different communication needs: low, moderate and intensive. The experimental results included in this paper are compared to random distributions of tasks across the entire environment, showing important improvements in terms of total execution time and amount of data transferred through the network.