Benchmarking financial database queries on a parallel machine

A database benchmark exercise on a parallel machine (a Kendall Square Research KSR1-32 virtual shared memory computer) for a large UK bank is discussed. The benchmark involved 8 queries running on a subset of the bank's database across 10 processors. All queries were of the MIS variety, and included "real" marketing queries. Some queries give speedup of over 5 times on 10 processors. As a result of the data distribution, most queries showed speedup for up to 5 processors but then no more speedup until 10 processors were used. Tests suggest that the limit to I/O throughput lies in the disk controllers. The effectiveness of query optimisation and decomposition are also discussed.<
>