[Gate-users] segfault ITK multithreader

[Simon Rit]

Hi,
Not in my experience. But reading again your first email, you should know
that by default, the number of threads is set to the number of cores. So
running 32 simulations on a 64 core system means 32*64 concurrent
threads... It might be that your system is not about it.
To test the benefit of multithreading, try running one simulation only and
increase ITK_GLOBAL_DEFAULT_NUMBER_OF_THREADS incrementally.
Simon

On Fri, Sep 29, 2017 at 4:03 PM, Clemens S. <clem.schmid at gmail.com> wrote:

> Dear Simon,
>
> thank you very much for your help! I recompiled ITK
> with ITK_USE_THREADPOOL=ON and it seems to run stable now.
>
> While compiling, I tried out the
> export ITK_GLOBAL_DEFAULT_NUMBER_OF_THREADS=1
> method. I haven't done serious benchmarks, but the simulations seem to
> finish much faster with that setting than with the threadpool. Could there
> be a lot of overhead with multithread and FFD?
>
> Anyway, thank you again for your help!
>
> Kind regards
> Clemens
>
> 2017-09-29 13:22 GMT+02:00 Simon Rit <simon.rit at creatis.insa-lyon.fr>:
>
>> Hi,
>> I'm not sure but I had similar issues at some point. I first fixed
>> something in ITK:
>> http://review.source.kitware.com/#/c/21937
>> Then the problem was a bit more explicit, it could not create enough
>> threads. I never explained it in my case (see:
>> https://public.kitware.com/pipermail/insight-users/2017-Janu
>> ary/054167.html) but turning on ITK_USE_THREADPOOL at compilation of ITK
>> fixed the problem. Can you turn on ITK_USE_THREADOOL with the latest ITK to
>> see if that solves your issue?
>> Note that you can always turn off multithreading using the env variable:
>> export ITK_GLOBAL_DEFAULT_NUMBER_OF_THREADS=1
>> Hope this helps,
>> Simon
>>
>> On Fri, Sep 29, 2017 at 12:14 PM, Clemens S. <clem.schmid at gmail.com>
>> wrote:
>>
>>> Dear fellow GATE users,
>>>
>>> I have a problem with running multiple parallel GATE simulations. After
>>> a while (around 10 minutes), simulations start segfaulting (stacktrace
>>> attached). This does not happen every or at the same time; sometimes all
>>> simulations finish correctly. It also happens if I don"t start as many
>>> simulations, but I feel not as frequently.
>>>
>>> I use the FixedForcedDetection Actor and start around 32 simulations at
>>> once on a 64 thread system. The simulation is attached.
>>>
>>> AFAIK, I did not enable multithread mode on any GATE-related program by
>>> hand when installing them.
>>>
>>> I am using
>>>  - GATE 7.2
>>>  - root 5.34/30
>>>  - simpleITK 4.10.1
>>>  - RTK 1.3.0
>>>
>>> Thank you for your help!
>>>
>>> Kind regards
>>> Clemens Schmid
>>>
>>> _______________________________________________
>>> Gate-users mailing list
>>> Gate-users at lists.opengatecollaboration.org
>>> http://lists.opengatecollaboration.org/mailman/listinfo/gate-users
>>>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.opengatecollaboration.org/pipermail/gate-users/attachments/20170929/c6d960e4/attachment-0001.html>