Abstract:In this paper, we present and evaluate algorithms to address combined path and server selection (CPSS) The evolution of the Internet and differentiated services has expanded the scope of the global information infrastructure and increased connectivity among service providers and clients requesting services for multimedia applications. As this infrastructure scales, service providers will need to replicate data and resources on the network to serve more concurrent clients. Efficient and adaptive resource management mechanisms are required to deal with highly dynamic environments (e.g. those that involve mobile clients and hosts). Quality of Service (QoS) routing techniques have been proposed to improve the network utilization by balancing the load among the individual network links. Server selection policies direct the user to the "best" server while statically treating the network path leading from the client to the server as pre-determined by the routing tables, even though there may exist multiple alternative paths. While the two techniques can independently achieve some degree of load balancing, we argue that in multimedia environments, where the applications are highly sensitive to QoS parameters like bandwidth and delay, high-level provisioning mechanisms are required to address the route selection and server selection problem in a unified way. Such integrated mechanisms can potentially achieve higher system-wide utilization, and therefore allow more concurrent users. Furthermore, in a highly dynamic environment, cost-effective QoS provisioning techniques for server and network resources must be able to tolerate some information imprecision and be able to work effectively with approximate system state information.In previous work, we have developed a model that addresses the combined routing and scheduling problem based on system residue capacities [FV99]. In this paper, we (a) develop and evaluate a family of policies for composite routing and scheduling and (b) test and understand the performance of the CPSS policies under different levels of information imprecision and varying traffic patterns.The rest of this paper is organized as follows. Section 2 describes the server and network model and presents the general case of the CPSS algorithm. In Section 3 we develop heuristics to solve the CPSS problem. Section 4 deals with the collection and update of state information using a directory service and discusses several policies for information update. Section 5 contains a performance evaluation of the CPSS policies and update mechanisms. Section 6 describes related work and concludes with future research directions.