Parallel execution of database queries in a shared nothing multiprocessor environment, needs control mechanisms which guarantee a correct global execution of each query. In the authors system, the control is integrated into the query code at compile time. According to the required schedule for the query, the compilation process combines a set of basic control operations into a control graph mapped on the query graph. This control which allows pipelining between data operations, achieves the dataflow control and enforces the schedule of program. Using fine grain control operations combined into control schemas makes the control flexible, as new strategies could be easily supported. In their prototype the parallel LERA language is used.
展开▼