Adaptive query scheduling for mixed database workloads with multiple objectives

Ideally, a data warehouse would be able to run multiple types of queries concurrently, meeting different performance objectives for each type. However, due to the difficulty of managing mixed workloads, most commercial systems segregate distinct workload components by using strict resource partitioning and/or time multiplexing. This approach avoids unexpected resource contention, but when one workload component does not fully use its allocated resources, those resources may then lie unused even if they could greatly improve the performance of another component. We focus here on adaptively scheduling mixed workloads that have multiple objectives. We use our experimental framework for testing policies to evaluate the extent to which prior approaches to adaptive workload scheduling address mixed workloads. Our experiments demonstrate the difficulty of searching for solutions in the space of scheduling dynamic mixed workloads. We discuss why prior approaches do not address certain scenarios and then demonstrate how leveraging additional knowledge would allow one approach to succeed, if that knowledge were available.