The number of stream-based information sources is increasing, so query processing on streams is getting more
important. Window-join is one of typical operations for stream data processing. It uses sliding-windows to keep recent input tuples, and generate the join result from the tuples in the window. To reduce processing cost of window-join, some parallel window-join algorithms have been proposed in the context of distributed environments. Their main objectives are increasing throughput. They do not concern about reducing processing delay, because delay caused network transfer is unavoidable in distributed environments. In our research, we focus on reducing processing delay of window-join by using multi-core environments instead of distributed environments. We propose and evaluate two parallel window-join algorithms for multiple input streams. The first one is Parallel Tree-based Join, which makes multiple cores execute multiple join trees with different join orders. Each join tree is optimized to minimize processing delay for tuples from one stream. Another is Parallel MJoin, which is an extension of existing MJoin algorithm. Parallel MJoin concurrently performs multiple probe processes instead of sequential probe used in original MJoin. This paper also includes experimental evaluation to investigate efficiency of two methods.
各種検索
サポート
ヘルプ