Performance Issues in Adaptive Query Processing Fred Reiss U.C. Berkeley Database Group
What Is an Eddy? Query Optimizer Operator Output Static Plan
Adaptive Plan Eddy What Is an Eddy? Query Operator Tuple Pool Routing Policy Output
Eddy Performance Agenda Goal: Replace the query optimizer Mechanism –Make adaptivity cheap Policy –Simple and effective adaptive routing policy
Mechanism Goal: make adaptivity cheap Minimize overhead –Use batching to amortize decisions Offset overhead –Use the tuple pool to cluster similar tuples
Policy Goal: simplicity –Want a routing policy that handles a wide variety of situations Randomized routing –Lottery scheduling –Machine learning / statistical models Hybrid/adaptive routing –Use a static optimizer as a subroutine