[Paraview] Fwd: Stream Tracer Scaling

[Berk Geveci]

Hi folks,

To close the loop on this issue, this was indeed a bug. It only happened
when running in parallel and using a cell based locator (which Amir needed
due to the nature of his data). The bug caused to computation time to
increase significantly as the number of seeds increased. It caused the
re-computation of the same spatial binning structure (most expansive part
of the algorithm) per seed. The fix was simply deleting a few lines of
unnecessary code. The biggest performance improved I ever achieved by
deleting code :-)

I am working on getting the fix incorporated into VTK and then ParaView. It
should take a week or two.

Best,
-berk



On Tue, Apr 22, 2014 at 9:07 AM, Berk Geveci <berk.geveci at kitware.com>wrote:

> Hmmm. It certainly should not be much slower. Even in the worst load
> balancing case, it should be at most slightly slower. I will take a look.
>
> -berk
>
>
> On Mon, Apr 21, 2014 at 6:20 PM, Amir A <amiramir12345 at outlook.com> wrote:
>
>> Hi Berk:
>>
>> Thank you for your reply. I am doing the visualization on a distributed
>> cluster. And the StreamTracer filter is used with a point source of radius
>> 0.1 and 100 seeds.
>>
>> It is surprising that the streamtracer in parallel takes orders of
>> magnitude more time than when applied in serial. I tried to first apply D3
>> filter and then the streamtracer. That helps a little bit but still the
>> time is prohibitively long (about an hour compared to ~5min when done in
>> serial). I do not know if I am missing something or this is usual.
>>
>> Here is the visualization that I want to create in parallel (I had to do
>> it in serial because of this problem):
>> https://www.dropbox.com/s/r3kkon3khjdpqs6/foam_3.png
>>
>> The original dataset is about 4GB.  Here is a smaller one (~1GB) with the
>> state files:
>> https://www.dropbox.com/s/r6lh3sx2wojfva2/sample.tar.gz
>>
>> I would highly appreciate any suggestions.
>> -Amir
>>
>> ------------------------------
>> From: berk.geveci at kitware.com
>> Date: Mon, 21 Apr 2014 11:34:25 -0400
>> To: andy.bauer at kitware.com
>> CC: paraview at paraview.org
>> Subject: Re: [Paraview] Fwd: Stream Tracer Scaling
>>
>>
>> Actually, if the integration is fairly load balanced across partitions,
>> the streamline integration can scale OK. It is highly dependent on how the
>> seeds are distributed and if the integration ends up clustering on a few
>> processes. Amir:
>>
>> - How big is your seed source? Does it span multiple partitions? How many
>> seed points?
>> - Do the streamlines stay together or do the move around the domain more
>> independently?
>>
>>  Are you running on a single machine by the way? I have been thinking
>> about parallelizing the streamline filter over threads, which should scale
>> better on a single machine.
>>
>> Best,
>> -berk
>>
>>
>> _______________________________________________ Powered by
>> www.kitware.com Visit other Kitware open-source projects at
>> http://www.kitware.com/opensource/opensource.html Please keep messages
>> on-topic and check the ParaView Wiki at:
>> http://paraview.org/Wiki/ParaView Follow this link to
>> subscribe/unsubscribe: http://www.paraview.org/mailman/listinfo/paraview
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.paraview.org/pipermail/paraview/attachments/20140429/37b6af3c/attachment.html>