Dynamic resource allocation strategy to handle hot-spots

Hot-spot is one of the main threats to Internet applications where non-deterministic workload is possible due to the open Internet architecture. Application replication is a promising approach to address such problem. In this paper, we investigate the selection strategy of rescue servers from the viewpoint of the heterogeneity of server capability. We first map the allocation of rescue servers into a graph problem. Then four heuristic functions are proposed to obtain approximate solutions. Detailed performance study of these algorithms shows that our ASA selection strategy, which is a greedy-based algorithm, performs the best when the overall system load and the replication latency among rescue servers are taken into consideration.