Parallel Computer Architecture

4.4 Illustrating Workload Characterization 243

our choice of problem size and number of processors for which we want to represent

this situation.

Summary

The preceding discussion shows that the results of an evaluation study can be mis-

leading if we don't cover the space adequately: we can easily choose a combination

of parameters and workloads that demonstrates good performance benefits from a

feature such as block transfer (for example, a relatively small problem size, big

caches, and a small cache block size), and we can just as easily choose a combina-

tion that doesn't. It is therefore very important that we incorporate sound method-

ologica ...