PAPI in the ports

Thomas Vogt thomas at
Thu Jun 14 11:47:44 UTC 2007


NOC Meganet wrote:
> On Thursday 14 June 2007 06:59:41 Thomas Vogt wrote:
>> Hi
>> Thats sounds nice. You wrote "The goal of the PmcTools project is to
>> provide FreeBSD's developers and system administrators with
>> non-intrusive, low-overhead and innovative ways of measuring and
>> analysing system performance" your website. Have you ever measured the
>> performance impact of such tools?
>> I'm interested to run such tool on production machines in the future but
>> only if the performance impact isn't that high.
> Hi
>  ... even if the tool (whatever it is) does not require much system resources 
> it triggers processes in order to measure their or the system's performance. 
> So long as you do not push the system to or over the edge you might not get 
> valuable numbers. So I mean, any performance measuring does or must stress 
> the system. That is at least my understanding. So probably running 
> performance tests on a production server is not the very best idea.

Yes you're right. But to have a such feature on a production system is
nice if you have to debug issues.  If I get a problem I can easily debug
it without to simulate everything in the lab. It's not that I want to
run such features 24/7.

But perhaps "options HWPMC_HOOKS" in the kernel enables some hooks which
could cause performance penalties even if I don't actively use such
feature very often. Like many other debug options in the kernel. I don't


More information about the freebsd-performance mailing list