Compression is an intuitive way to boost the performance of a database system. However, compared with other physical database design techniques, compression consumes large amount of CPU power. There is a trade-off bet...Compression is an intuitive way to boost the performance of a database system. However, compared with other physical database design techniques, compression consumes large amount of CPU power. There is a trade-off between the re- duction of disk access and the overhead of CPU processing. Automatic design and adaptive administration of database systems are widely demanded, and the automatic selection of compression schema to compromise the trade-off is very important. In this paper, we present a model with novel techniques to integrate a rapidly convergent agent-based evolution framework, i.e. the SWAF (SWarm Algorithm Framework), into adaptive attribute compression for relational database. The model evolutionally consults statistics of CPU load and IO bandwidth to select compression schemas considering both aspects of the trade-off. We have im- plemented a prototype model on Oscar RDBMS with experiments highlighting the correctness and efficiency of our techniques.展开更多
基金Project (No. 2004AA4Z3010) supported by the National Hi-Tech Research and Development Program (863) of China
文摘Compression is an intuitive way to boost the performance of a database system. However, compared with other physical database design techniques, compression consumes large amount of CPU power. There is a trade-off between the re- duction of disk access and the overhead of CPU processing. Automatic design and adaptive administration of database systems are widely demanded, and the automatic selection of compression schema to compromise the trade-off is very important. In this paper, we present a model with novel techniques to integrate a rapidly convergent agent-based evolution framework, i.e. the SWAF (SWarm Algorithm Framework), into adaptive attribute compression for relational database. The model evolutionally consults statistics of CPU load and IO bandwidth to select compression schemas considering both aspects of the trade-off. We have im- plemented a prototype model on Oscar RDBMS with experiments highlighting the correctness and efficiency of our techniques.
