Date of this Version
Cluster Computing has emerged as a new paradigm for solving large-scale problems. To enhance QoS and provide performance guarantees in cluster computing environments, various workload models and real-time scheduling algorithms have been investigated. The divisible load model, propagated by divisible load theory, models computations that can be arbitrarily divided into independent pieces and provides a good approximation of many real-world applications. However, researchers have not yet investigated the problem of providing performance guarantees to divisible load applications. Two contributions are made in this paper: (1) divisible load theory is extended to compute the minimum number of processors required to meet an application’s deadline; and (2) the first cluster-based, real-time scheduling algorithm designed specifically for arbitrarily divisible loads is presented and evaluated.