[petsc-users] Obtaining bytes per second

[Matthew Knepley]

On Thu, May 7, 2015 at 9:23 AM, Justin Chang <jychang48 at gmail.com> wrote:

> So to summarize, if I understand everything, I should do the following:
>
> 1) calculate the flop/byte ratio for various problem sizes and solver
> methods on one process and:
>

Should be roughly invariant to problem size.


> a) use the ratio to estimate the ideal flops/s based on the stream bw.
> Compare this to the measured flops/s from my Petsc program?
>

Yes


> b) show the associated wall-clock times.
>

Your strong scaling plots are fine here.


> c) warn that this model is uncached and merely gives a rough estimate of
> the performance.
>

Yes, I would say "does not deal with cache effects".


> 2) Do a strong-scaling/speed up study to illustrate how these problems
> scale across multiple processes. Optionally see what the optimal number of
> processes is required for various problem sizes and solvers.
>
> Am I missing anything?
>

Sounds good to me.

   Matt


> Thanks,
> Justin
>
>
> On Thursday, May 7, 2015, Matthew Knepley <knepley at gmail.com> wrote:
>
>> On Thu, May 7, 2015 at 8:35 AM, Jed Brown <jed at jedbrown.org> wrote:
>>
>>> Matthew Knepley <knepley at gmail.com> writes:
>>> > I think it would be much more interesting, and no more work to
>>> >
>>> >   a) Model the flop/byte \beta ratio simply
>>> >
>>> >   b) Report how close you get to the max performance given \beta on
>>> your
>>> > machine
>>>
>>> This is also easily gamed; just use high-memory data structures, extra
>>> STREAM copies, etc.
>>>
>>
>> I completely agree that all the performance measures can be gamed, which
>> is why
>> its important to always show time to solution as well.
>>
>>    Matt
>>
>> --
>> What most experimenters take for granted before they begin their
>> experiments is infinitely more interesting than any results to which their
>> experiments lead.
>> -- Norbert Wiener
>>
>


-- 
What most experimenters take for granted before they begin their
experiments is infinitely more interesting than any results to which their
experiments lead.
-- Norbert Wiener
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/petsc-users/attachments/20150507/e0c2f1e6/attachment.html>