Analysis of Cluster Topologies for Dynamic Workload Balancing Strategies in Image Databases

摘要

A search for objects or image components in a system for the archival and retrieval of images goes beyond the means of static databases and requires dynamic feature extraction. Since the computational and storage requirements exceed the possibilities of computer architectures with a single processing element, parallel architectures ― in this case cluster architectures ― have to be considered. The transfer and processing is shared by a number of nodes and the reasonable price per node enables the design of large-scale systems. However, the choice of an adequate cluster organisation is not straightforward due to the specific demands of an image database application. This paper investigates different cluster arrangements and proposes a platform which supports the implementation of dynamic retrieval strategies in existing systems. A particular emphasis is placed on the workload balancing across the cluster.