Problems for Resource Brokering in Large and Dynamic Grid Environments

Abstract. Running workloads in a Grid environment may become a challenging problem when no appropriate means are available for resource brokering. Many times resources are provided under various administrative policies and agreements that must be known in order to perform adequate scheduling decisions. Thus, providing suitable solutions for resource management is important if we want to cope with the increased scale and complexity of such distributed system. In this paper we explore the key requirements a brokering infrastructure must meet in large and dynamic Grid environments and illustrate how these requirements are addressed by a specialized infrastructure, DI-GRUBER- a distributed usage service level agreement (uSLA) brokering service. The accuracy function of the brokering infrastructure connectivity and the performance gains when a client scheduling policy is employed are analyzed in high detail. In addition, a performance comparison with a P2P-based distributed lookup service is performed to illustrate the performance differences between two different technologies that address similar problems (Grids that focus on federated resource sharing scenarios and P2Ps that focus on self-organizing distributed resource sharing systems, in which most of the communication is symmetric). 1