[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-users] Why do kflops and mips vary so much?

On Thu, Apr 22, 2010 at 1:47 PM, Matt Hope <Matt.Hope@xxxxxxxxxxxxxxx> wrote:

That's a difference of 1% I wouldn't expect better accuracy from a benchmark in an uncontrolled environment (i.e. other process could be running at the same time on a non realtime OS)

The noise and the inherent difficulties in really measuring the "performance" of a machine in an uncontrolled environment are why I'm asking: what is it doing to get these numbers? Why is there any noise at all?

Anything that runs long enough to smooth out effects from things like dynamic power levels or turbo boost will end up being subject to the vagaries of the OS time slicing, even on a relatively uncontended system. If other things are running as well then even more so.

Read: the numbers are mostly useless. :)

I'm all for running things once at startup and then never again. Because I think the numbers don't mean a whole lot and because I'd rather Condor just get out of the way if there's a slot not doing anything, not benchmark my machine.

- Ian