An Evaluation of Architectural Alternatives for Rapidly GrowingDatasets: Active Disks, Clusters, SMPs

Growth and usage trends for several large datasets indicate that there is aneed for architectures that scale the processing power as the datasetincreases. In this paper, we evaluate three architectural alternatives forrapidly growing and frequently reprocessed datasets: active disks, clusters,and shared memory multiprocessors (SMPs). The focus of this evaluation is toidentify potential bottlenecks in each of the alternative architectures and todetermine the performance of these architectures for the applications ofinterest. We evaluate these architectural alternatives using a detailedsimulator and a suite of nine applications. Our results indicate that for mostof these applications Active Disk and cluster configurations were able toachieve significantly better performance than SMP configurations. Active Diskconfigurations were able to match (and in some cases improve upon) theperformance of commodity cluster configurations.