dolfin team mailing list archive

[Johannes Ring <johannr@xxxxxxxxx>]

On Tue, Jun 29, 2010 at 1:03 PM, Johannes Ring <johannr@xxxxxxxxx> wrote:
> On Tue, Jun 29, 2010 at 12:41 PM, Anders Logg <logg@xxxxxxxxx> wrote:
>> On Tue, Jun 29, 2010 at 12:34:48PM +0200, Johannes Ring wrote:
>>> On Tue, Jun 29, 2010 at 11:16 AM, Anders Logg <logg@xxxxxxxxx> wrote:
>>> > On Tue, Jun 29, 2010 at 11:09:47AM +0200, Johannes Ring wrote:
>>> >> I backported this benchmark to 0.9.7 and it ran for over 17 minutes.
>>> >
>>> > !
>>> >
>>> > That's quite a good speedup. :-)
>>> >
>>> >> On DOLFIN-dev it run for about 55 seconds. Maybe we should reduce
>>> >> the
>>> >> problem size somewhat before I backport further?
>>> >
>>> > Sure, but not too much. I think we can improve it further (by a factor
>>> > 10 or so) and then it might run too quickly. I also suspect DOLFIN
>>> > 0.9.7 to be the slowest (or one of the slowest).
>>> >
>>> > Try reducing SIZE by a factor 2 or so.
>>>
>>> That reduced the runtime to just under nine minutes for 0.9.7 and 33
>>> seconds for bzr tip. I also tried DOLFIN 0.7.3 and it is as slow as
>>> 0.9.7.
>>
>> Interesting. But I guess it's not a problem since you only have to run
>> it once to get the numbers.
>
> Yes, that's right. I have reduced the size and will backport the benchmark.

The web page has been updated with the new benchmark now:

  http://www.fenics.org/bench/

Johannes

> Johannes
>
>> Btw, I compared running with and without progress bar in a simple loop
>> that just does one addition per loop (like in the benchmark) and the
>> difference was a factor 3. Not sure if that is reasonable or not. It
>> depends on how much a function call should cost compared to adding two
>> numbers.
>>
>> --
>> Anders
>>
>> -----BEGIN PGP SIGNATURE-----
>> Version: GnuPG v1.4.10 (GNU/Linux)
>>
>> iEYEARECAAYFAkwpzekACgkQTuwUCDsYZdGQegCfeiQkZFx37Ejd7EnTL7edttJQ
>> q/wAnjyLAr7nsCzSDXWpI+fR1yhjSaXO
>> =sOx2
>> -----END PGP SIGNATURE-----
>>
>>
>