Adaptive Load Sharing for Clustered Digital Library Servers

This paper investigates load balancing strategies for clustered Alexandriadigital library (ADL) servers. The ADL system, which provides on-lineinformation searching and browsing of spatially-referenced materials throughthe World Wide Web, involves intensive database I/O and heterogeneous CPUactivities. Clustering servers can improve the scalability of the ADL systemin response to a large number of simultaneous access requests. One difficultyaddressed is that clustered workstation nodes may be non-uniform in terms ofCPU and I/O speeds. An optimization scheme is proposed in this paper todynamically monitor the resource availability, use a low-cost communicationstrategy for updating load information among nodes, and schedule requests basedon both I/O and computation load indices. Since the accurate cost estimationfor processing database-searching requests is difficult, a sampling andprediction scheme is used to identify the relative efficiency of nodes forsatisfying I/O and CPU demands of these requests. A set of experiments usingthe ADL traces have been conducted to verify the effectiveness of the proposedstrategies.