The implementation and performance of compressed databases

In this paper, we show how compression can be integrated into a relational database system. Specifically, we describe how the storage manager, the que ry execution engine, and the query optimizer of a database system can be ex tended to deal with compressed data. Our main result is that compression ca n significantly improve the response time of queries if very light-weight c ompression techniques are used. We will present such light-weight compressi on techniques and give the results of running the TPC-D benchmark on a so c ompressed database and a non-compressed database using the AODB database sy stem, an experimental database system that was developed at the Universitie s of Mannheim and Passau. Our benchmark results demonstrate that compressio n indeed offers high performance gains (up to 50%) for IO-intensive queries and moderate gains for CPU-intensive queries. Compression can, however, al so increase the running time of certain update operations. In all, we recom mend to extend today's database systems with lightweight compression techni ques and to make extensive use of this feature.