I've recently taken it upon myself to try and get a better idea of how the machines we have here in the College compare to the other kinds of iron you can buy these days. (Part of the impetus behind this was the Alpha project: new, fast machines make you want to know how much better they are than your old, slower ones). We think we have a pretty nice setup here at CCS, and while a lot of that's due to software, it'd be nice to know how our machines really perform.
To that end, I've (just) started running a series of benchmarks
on our various architectures. Some are standards, like the dhrystone
benchmark, others are slightly more obscure. lmbench-1.0, written by
Larry McVoy (lm@sgi.com) may well prove to be the most interesting.
At the moment I've only got some very preliminary
results; hopefully we'll get a more complete view as I have time
to run the tests and compile a real report.