Thomas M. Comeau
STScI
Victoria Park
STScI
The Data Archive and Distribution System (DADS) manages the Hubble Data Archive (HDA), a WORM Optical Disk Archive which contains over two terabytes of Hubble Space Telescope data. One fortunate side effect of retrievals from the HDA is that all retrieval requests are permanently logged in database tables. Queries against these tables provide a complete history of requests serviced by DADS. Throughput of the archive is an important bottleneck for overall DADS performance. If a significant fraction of retrieve requests could be serviced from a cache, this bottleneck would be bypassed. Additionally, retrieve requests currently interfere with the process of ingesting new datasets. This is a potentially serious problem in the post Servicing Mission 97 (SM97) era, when ingest rates are close to the limit of this same archive bottleneck. We describe a system which uses the database request logs as input to a flexible cache model. The model permits changes to the size of the cache, replacment strategy, and preloading the cache. We also discuss possible cache sizes and replacement strategies, and their effect on DADS performance in the post-SM97 system.