Benchmarks

My company has been going through a process of evaluating Unix servers which will be used to run a moderately large Oracle database system. The time for a decision is drawing near, and there is still considerable anxiety about it. Most of that anxiety revolves around the available benchmarks for the servers, and how to estimate their meaning.

For example, we have a set of "tps", TPC-A and TPC-C benchmarks for the three servers in question (a Data General AViiON 8500, HP H50, and Sun SPARCserver 1000). I'm not sure what to make of the "tps" numbers, as they are sometimes defined as estimates, and sometimes as TPC-A figures. As for the others, they are generally in what looks to us like the same ballpark.

How much variation can one expect between the numbers seen in one of these benchmarks, and what can be expected in real life? Is there a generally accepted range, such as "Real life is within 10% one way or the other of TPC-A numbers"?

Any ideas about how to look at these benchmarks would be much appreciated.

Also, if anyone has actually compared two or more of the systems mentioned, I'd love hear how your compared them and what the results you saw were.

I'd appreciate any responses either here or through email.

Thanks.

-- 
Joe Halpin 
jhalpin_at_netcom.com
---------------------------------------------------------------------------