A graph-based approach to action scheduling in a parallel database system

Parallel database machines are meant to obtain high performance in transaction processing, both in terms of response time adn throughput. To obtain high performance, a good scheduling of the execution of the various actions in transactions is crucial. This paper describes a graph-based technique for the scheduling of actions that allows both intra- and inter-transaction parallelism. The technique has a number of properties that distinguish it from other approaches. In the first place, the scheduling is based on high-level graph processing, thus enabling clear semantics, parallelism in the scheduling task, and low over-head