Load balancing and query optimization in dataflow parallel evaluation of Datalog programs

A dataflow model to obtain parallelism in the evaluation of Datalog is presented. This model performs query evaluation as a dataflow through a network of communicating concurrent processes capable of solving the query. This process network is based on the intensional database definition, plus the concrete query to be evaluated. A cost model to cope with the load balancing problem is described. A load balancing algorithm is presented and discussed. An algorithm to optimize the evaluation is described which is based on process network rewriting. This utilizes information in the query bindings to be evaluated in order to optimize the dataflow graph.