Execution plan balancing

A novel relational query optimization technique for use in shared memory multiprocessor database systems is described. A collection of practical algorithms for allocating computational resources to parallel select-project filter (SPJ) query execution plans is presented. The computational resources considered include disk bandwidth, memory buffers and general-purpose processors. The goal of the allocation algorithms is to produce minimum duration execution strategies with computational resource requirements that are less than the given system bounds. Preliminary experimental results indicate that the algorithms can be realized and are effective in producing good execution plans. Disk bandwidth appears to be the critical system resource. The most effective means to decrease complex query response time appears to be by reducing disk contention. This can be achieved by increasing the total number of disks and/or rearranging the placement of data on disks