Automatic Workload Management for Enterprise Data Warehouses.
Abhay Mehta, Chetan Gupta, Song Wang, Umeshwar Dayal · IEEE Data(base) Engineering Bulletin · 2008
Modern enterprise data warehouses have complex workloads that are notoriously difficult to manage. Additionally, RDBMSs have many “knobs” for managing workloads efficiently. These knobs affect the performance of query workloads in complex interrelated ways and require expert manual attention to change. It often takes a long time for a performance expert to get enough experience with a large warehouse to be able to set the knobs optimally. Typically the warehouse and its workload change significantly within that time. This makes the task of manually optimizing the knob settings on a warehouse an impossible one. In this context, our goal is to create self managing Enterprise Data Warehouses. In this paper we describe some recent advances in building an automatic workload management system. We test this system against real workloads against real enterprise data warehouses.