Strategies of Domain Decomposition to Partition Mesh-Based Applications onto Computational Grids

In this paper, we evaluate strategies of domain decomposition in Grid environment to solve mesh-based applications. We compare the balanced distribution strategy with unbalanced distribution strategies. While the former is a common strategy in homogenous computing environment (e.g. parallel computers), it presents some problems due to communication latency in Grid environments. Unbalanced decomposition strategies consist of assigning less workload to processors responsible for sending updates outside the host. The results obtained in Grid environments show that unbalanced distributions strategies improve the expected execution time of mesh-based applications by up to 53%. However, this is not true when the number of processors devoted to communication exceeds the number of processors devoted to calculation in the host. To solve this problem we propose a new unbalanced distribution strategy that improves the expected execution time up to 43%. We analyze the influence of the communication patterns on execution times using the Dimemas simulator.