Randomised Block Size Scheduling Strategy for Cluster-Based Image Databases

摘要

Systems for the archival and retrieval of images are used in many areas and depict an essential module, for general multimedia databases. However, the computational demands of the retrieval operations often surpass the capabilities of traditional architectures, thus the integration of concepts for parallel and distributed storage and processing is required. Cluster architectures offer advantages compared to other parallel architectures, as the transfer, storage, and processing effort is shared by a number of independent nodes. The image distribution over the available nodes lias a decisive impact on the speedup and efficiency, thus suit-able stmtegi.es for workload balancing are needed in order to improve the overall performance. This paper discusses, evaluates and compares two strategies called. LTF (Largest Task First) and RBS (Randomised Block Size) for dynamic workload, balancing across an image retrieval cluster. They consider the large memory requirements of multimedia data and minimise the communication between the nodes. Performance measurements resulted in a significant improvement of the system response time.