topic Re: a simple performance check program in Operating System - HP-UX

a simple performance check program

TwoProc — Wed, 29 Sep 2010 18:45:38 GMT

I need a simple c program that I can compile across hpux on parisc and hpux on Itanium to do some simple cpu tests for speed. And, I need roughly the same to check for memory throughput, or both at the same time.

For example, I know of Bonnie for testing disk I/O as a very simple C program I can compile on either platform. Can anyone point me to one for HPUX for Itanium vs PARISC for testing CPU (integer, floating pt, etc), and memory allocation, de-allocation, reading and writing?
If not a single program, then multiple programs for testing each area would be fine as well.

I could write something relatively easily, but I have to believe this has been done already.

Thank you!

John

Re: a simple performance check program

TwoProc — Wed, 29 Sep 2010 20:33:59 GMT

I found a copy of the old Byte magazine program nbench converted to Linux. I altered the makefile a bit, and I got what I needed.

Re: a simple performance check program

TwoProc — Wed, 29 Sep 2010 20:35:16 GMT

I don't know why the system gave this question a rabbit with no responses from the group and no pts awarded. Seems goofy.

Re: a simple performance check program

rick jones — Wed, 29 Sep 2010 22:49:13 GMT

The problem with simple benchmarks is the ease with which they can become simplistic. And the more simplistic, the less accurate a prediction for "real" applications. For example, something measuring CPU with a really tiny working set may fit entirely not just in the last level cache, but also in the lower level caches. In those cases, frequency can "win" and show a delta that will be rather different from real applications.

The gold standard for measuring CPU performance would include SPEC's CPU2006 suite. That though is probably out of the reach of many.

For an explicit memory bandwidth microbenchmark, McCalpin's STREAM benchmark is widely accepted.

Re: a simple performance check program

Dennis Handly — Thu, 30 Sep 2010 05:35:55 GMT

>I don't know why the system gave this question a rabbit with no responses from the group

You get a rabbit if you assign 8 or more to any reply. You also get it if you close it and say that the problem was resolved or some such wording.

Re: a simple performance check program

TwoProc — Thu, 30 Sep 2010 19:44:09 GMT

yeah Rick, I agree - the problem is that I am benchmarking with regression testing, and I'm not seeing what I'm supposed to be seeing in a total throughput capacity, at least not even close to what I'm expecting. So, I wanted to see that, at the very least, basic cpu computations and memory allocations, etc are indeed faster with the newer cpu systems and hardware than the old one. Then, I could know that it should be a tuning/setup issue, and not a pure horsepower cpu/memory issue.

I have the big gorrilla comparison, at that point all I wanted the small one.

Thanks for the helpful advice guys.