Parallel Computer Architecture

4.1 Scaling Workloads and Machines 207

1.

Under what constraints should the problem be scaled? To define a scaling model,

some property must be kept fixed as the machine scales. These properties might

include data set size, memory usage per processor, execution time, number of

transactions executed per second, and number of particles or rows of a matrix

assigned to each processor.

2.

How should the problem be scaled? That is, how should the parameters in the

problem size vector

V

be changed to meet the chosen constraints?

To simplify the discussion, we begin by pretending that the problem size is deter-

mined by a single parameter n and examin ...