Optimal Balanced Assignments and a Parallel Database Application

In parallel database systems, distribution of the data among the processors has a signiicant impact on the response time and throughput of the system. The beneets of parallelism (using multiple processors to execute a query) must be balanced against its costs (communication, startup, and termination overhead). We formalize the problem of minimizing overhead while partitioning data uniformly across the processors. We derive lower bounds on these combinatorial problems and demonstrate how processors may be optimally assigned so as to achieve these lower bounds for a number of problem classes.