
Appendix:
Parallel Benchmark Suites
Despite the difficulty of identifying "representative" parallel applications and the
immaturity of parallel software (languages and programming environments), the
quantitative workload-driven evaluation of machines and architectural ideas must
go on. To this end, several suites of parallel "benchmarks" have been developed and
distributed. The term benchmarks is in quotations because of the difficulty of relying
on a single suite of programs to make definitive performance claims in parallel com-
puting. Benchmark suites for multiprocessors vary in the kinds of application
domains and run-time characteristic ...