Dynamic Action Scheduling in a Parallel Database System

This paper describes a scheduling technique for parallel database systems to obtain high performance, both in terms of response time and throughput. The technique enables both intra- and inter-transaction parallelism while controlling concurrency between transactions correctly. Scheduling is performed dynamically at transaction execution time, taking into account dynamic aspects of the execution and allowing parallelism between the scheduling and transaction execution processes. The technique has a solid conceptual background, based on a simple graph-based approach. The usability and effectiveness of the technique are demonstrated by implementation in and measurements on the parallel PRISMA database system.