Report ID
1998-06
Report Authors
Anurag Acharya, Mustafa Uysal, and Joel Saltz
Report Date
Abstract
Several application and technology trends indicate that it might be bothprofitable and feasible to move computation closer to the data that itprocesses. In this paper, we evaluate Active Diskarchitectures which integrate significant processing power and memory into adisk drive and allow application-specific code to be downloaded and executed onthe data that is being read from (written to) disk. The key idea is to offloadbulk of the processing to the disk-resident processors and to use the hostprocessor primarily for coordination, scheduling and combination of resultsfrom individual disks. To program Active Disks, we propose a stream-basedprogramming model which allows disklets to be executed efficiently and safely.Simulation results for a suite of seven algorithms from three applicationdomains (commercial data warehouses, image processing and satellite dataprocessing) indicate that for these algorithms, Active Disks outperformconventional-disk architectures.
Document
1998-06.ps643.62 KB