Efficient Execution of Multiple Queries on Deep Memory Hierarchy

This paper proposes a complementary novel idea, called MiniTasking tofurther reduce the number of cache misses by improving the data \ittemporal locality for \it multiple concurrent queries. Our idea isbased on the observation that, in many workloads such as decisionsupport systems (DSS), there is usually significant amount of datasharing among different concurrent queries. MiniTasking exploits suchdata sharing to improve data temporal locality by scheduling queryexecution at three levels: query level batching, operator level groupingand mini-task level scheduling.The experimental results with various types of concurrent TPC-H queryworkloads show that, with the traditional N-ary Storage Model (NSM)layout, MiniTasking significantly reduces the L2 cache misses by up to83\%, and thereby achieves 24\% reduction in execution time. With thePartition Attributes Across (PAX) layout, MiniTasking further reducesthe cache misses by 65\% and the execution time by 9\%. For the TPC-Hthroughput test workload, MiniTasking improves the end performance up to20\%.