[openib-general] Re: cycles_to_units is incorrect in rdma_lat, rdma_bw.

[Grant Grundler]

On Wed, Jun 15, 2005 at 06:16:53PM +0300, Michael S. Tsirkin wrote:
> Quoting r. Grant Grundler <iod00d at hp.com>:
> > Patch below adds "-g" (gettimeofday) and "-M=<CPU Mhz>" parameters
> > in case someone's firmware is less accurate than HP's.
> 
> I dont think we shall care about that until this comes about.

Shirley Ma (IBM) has asserted that was the case for machines
she was testing on. That's what started this email thread.

Shirley, did the patch to get_clock.c fix the problem?

> Maybe there shall be a separate utility to compare
> get_cycles and gettimeofday results, may be useful for debug.
> 
> Anyway, if get_cpu_mhz isnt reliable people can just dump raw
> cycles data and do the math outside the tool.

Yeah, that's a valid alternative. And the extra code doesn't make
this a better example. OTOH, it's more convenient to have the
math integrated in the test - one less step that can introduce errors.
Any one else have an opinion? Shirley?

thanks,
grant