On the design, implementation, and evaluation of a portable parallel database system

Abstract

The authors describe a complete implementation and evaluation of a parallel relational database system. The described implementation exploits parallel algorithms initially proposed for a hypercube to achieve its speedup. Specifically, attention is given to a portable parallel database system that exploits both parallel algorithms and data parallelism to expedite database processing. Two join algorithms are evaluated. It is shown that, for joins with a comparable number of tuples in each of the two joining relations, a bucket-based approach is preferable. However, if the two relations greatly differ in size, a broadcast-based approach is preferred.<>