Adaptive Load Balancing for Divide-and-Conquer Grid Applications

Divide-and-conquer has been demonstrated as a simple and efficient programming model for grid applications. In previous work, we have presented the divide-and-conquer based Satin system and its load balancing algorithm, clusteraware work stealing (CRS). In this paper, we provide a detailed analysis of CRS with respect to important properties of grid systems, namely scalability, heterogeneous compute clusters, and dynamically changing network interconnects. Our results show that CRS automatically adapts both to heterogeneous processor speeds and varying network performance, resulting in efficient utilization of the computing resources.