| dc.description.abstract | Query optimization is a quintessential element of modern Database Management Systems(DBMSs). Compile-time driven estimates and heuristics aid the compiler in selecting what is deemed the lowest cost Access Plan for a given query. These access plans are seldom optimal, and can oftentimes lead to under-performing query runtimes, with varying severity. Traditionally, domain experts painstakingly examine the access plans to detect and fix problem patterns. DistGALO, the successor to the previous GALO system, was developed to remedy this manual labour by incorporating a cluster of nodes to learn problem patterns in a distributed fashion and apply the fixes automatically. Several partitioning and pruning strategies are employed, including the RSACE module which gives user fine-grained control for trading off runtime versus template creation. In the experimental validation, DistGALO demonstrates the efficiency boost over our previous system using the synthetic TPC-DS benchmark and the effectiveness of the various pruning strategies. | en |
