4-022
To Compress or Not to Compress: Objectivity/DB Compressed Database Support
Andrew Hanushevsky/SLAC
The BaBar database, based upon the Objectivity/DB, has been in production since early 1999, and easily accommodates over 200TB per year. However, with increased luminosity and changes in the physics requirements, the amount of data that will be stored in Objectivity/DB will increase by almost an order of magnitude. At the proposed data rate, the amount of data stored in Objectivity/DB will exceed Moore’s Law on disk space price/performance. The potential added cost will significantly limit large-scale analysis of detector data and effectively reduce the detector luminosity; making increases in luminosity rather a moot point. One possible solution is to directly process compressed databases. While compression trades disk space for CPU resources, the trade-off may be worthwhile in certain circumstances. I discuss compressed database support in Objectivity/DB; including architecture, compression methods, compression ratios based on actual BaBar data, time-space tradeoffs, and when such trade-offs are reasonable.
Keywords:	OO Database, Objectivity, Scaling, Compression
Contact:	Mr. Andrew Hanushevsky
	Stanford Linear Accelerator Laboratory
	abh@stanford.edu