Minimizing latency and memory in DSMS
Yijian Bai, Carlo Zaniolo · 2008
Data Stream Management Systems (DSMSs) must support optimized execution scheduling of multiple continuous queries on massive, and frequently bursty, data streams. Previous approaches on optimizing memory consumption or response time (i.e., latency) usually produce very different algorithms. In this paper, we extend the popular chart-partitioning procedure, which was previously used for memory optimization on simple operator paths, to minimize latency as well as memory on complex query-graphs with tuple-sharing forks. Furthermore, we test the performance of algorithms that only assume knowledge of the average behavior of tuples and operators, against a theoretical one that assumes detailed knowledge on the behavior of individual tuples. These experiments show that the practical algorithms closely approximate the performance of the optimal ones.