Abstract

Compression is typically used for databases that have grown large enough to create a strain on system storage capacity. We argue here that database compression is attractive from a query processing viewpoint also and should therefore be implemented even when disk storage is plentiful. We study the compression ratio and query processing performance of a variety of compression algorithms, for different compression granularities, on a set of relations drawn from real world databases. Our study shows that attribute level compression is the best from a query processing perspective but has poor compression ratio. We then present a modified attribute level compression algorithm, based on non-adaptive arithmetic compression, called COLA, which simultaneously provides good query processing and reasonable compression ratios. We also analyze, for a range of relational queries, the performance benefits that COLA could be expected to provide. 1 Introduction Many database management systems provide...

Citations