Evaluation of Active Disks for Large Decision Support Databases

Growth and usage trends for large decision support databases indicate thatthere is a need for architectures that scale the processing power as thedataset grows. To meet this need, several researchers have recently proposedActive Disk architectures which integrate substantial processing power andmemory into disk units. In this paper, we evaluate Active Disks for decisionsupport databases. First, we compare the performance of Active Disks with thatof existing scalable server architectures: SMP-based conventional disk farmsand commodity clusters of PCs. Second, we evaluate the impact of severaldesign choices on the performance of Active Disks. We focus on the performanceimpact of interconnect bandwidth, amount of disk memory, disk-to-diskcommunication architecture and the host processor speed on decision supportworkloads. Our results show that for identical disks, number of processors andI/O interconnect, Active Disks provide better price/performance than bothSMP-based conventional disk farms and commodity clusters. Experimentsevaluating the impact of design alternatives in Active Disk architecturesindicate that: (1) for configurations up to 64 disks, a dual fibre channelarbitrated loop interconnect is sufficient even for the mostcommunication-intensive decision support tasks; (2) most decision support tasksdo not require a large amount of memory; (3) direct disk-to-disk communicationis necessary for achieving good performance on tasks that repartition all (or alarge fraction of) their dataset; and (4) the host processor is rarely abottleneck for configurations up to 128 disks.