Efficient Execution of Malleable Applications in Desktop Grids Using Credit Damping.

A desktop grid federation has enabled the organizations to solve large-scale game tree search applications via resource sharing. Since such applications can generate/prune tasks dynamically, the simple credit-based resource broker can easily lead to resource thrashing, causing a high computing overhead. We develop a credit damping component that is added between the broker and the credit database. By applying certain equalization function on the credit data, the broker avoids rapid interact to credit updates, therefore prevents thrashing and then enhances the system performance.