COREDB is a parallel database computer with hypercube topology. Currently, COREDB consists a cube manager and eight node computers, each of which contains a computation module and a communication module. Attached to each node computer is a secondary storage system for parallel database operations.
The benchmark method is widely used to measure performance of database systems. The benchmark method can only give us systemwide elasped time. We need, however, information of node computers such as elasped time and message traffic to represent performance of a parallel database computer in terms of various system parameters. This thesis develops a methodology and an environment realizing the methodology for measuring performance of a parallel database computer. The methodology employs combination of the benchmark method and the system monitor method. A system monitor measures performance parameters of parallel database systems which cannot be measured explicitly by a benchmark. Thus, combination of these two methods enables us to measure performance of the parallel database computer COREDB.
Though the Wisconsin benchmark is commonly used to measure performance of various database systems, it has some restrictions for appling to a parallel database computer. Therefore, we design and implement a new benchmark for measuring performance of the computer. We also design and implement a system monitor for collecting and analyzing performance parameters for each node computer. By combining these two implementations, we develop an environment for the measurement of parallel database system performance.
