Parallel Computer Architecture

262 CHAPTER 4 Workload-Driven Evaluation

4.5 CONCLUDING REMARKS

We now have a good understanding of the major issues in workload-driven evalua-

tion for multiprocessors: choosing workloads, scaling problems and machines, deal-

ing with the large parameter space, and choosing metrics. For each issue, we have a

set of guidelines and steps to follow, an understanding of how to avoid pitfalls, and a

means to understand the limitations of investigations. We also have a basis for our

own quantitative illustration of architectural trade-offs in the rest of the book. The

experiments in the book illustrate important points rather than evaluate trade-off ...