The CAD programs in our benchmark suite are a very important aid to
system designers today. Ideally, the designers would like to use these
programs interactively. Therefore, response time seems to be the
natural performance criterion. But, we chose to use the user time
(system + CPU) as our performance measure because the machines that we
were benchmarking were all compute servers and the response time is
heavily dependent on the load. Since the loading of these machines was
not under our control, we chose the next best performance criterion
i.e. user time. It seems reasonable to assume that these times will be
very close to the response times for personal machines. We also
limited the size of the examples we ran our programs on to those which
could fit into the main memory of all machines to make the user time
even closer to the response time.