Parallel Database Systems: The Future of High Performance Database Systems, DeWitt & Gray

This paper advocates the use of shared-nothing parallel database systems made from fast, inexpensive microprocessors. It gives examples of actual multiprocessors that follow this model. Because of the lower hardware complexity and network traffic, the shared-nothing architecture allows for a greater number of processors within the system and near-linear speedups and scaleups for OLTP loads. The paper argues that shared-nothing systems with a large number of processors are ideal for database applications because SQL code is easily executed in parallel, unlike most old software. The paper mentions data partitioning as a good way to improve speedup and scaleup, an idea which is further examined in the Google MapReduce paper.