Hi all, Just to follow up, the fix for this is now in the latest master. Cheers, Dave
On 12 Jul 2016, at 08:50, David Moxey <d.moxey@imperial.ac.uk> wrote:
Hi Fabian,
This is a bit of an irritating bug since, as you say, it _can_ work sometimes. Let me try to briefly explain the bug to give context and why you might be seeing this work fine on your laptop but not the cluster. It is basically due to the boundary conditions.
Due to a recent optimisation we needed to add a small communication call into a routine we use to extract edge values from elements, which is used in the boundary conditions to calculate the necessary values for the Riemann solver.
When running in parallel, we partition the mesh and the boundaries of the domain therefore lie on different processors. Depending on how many processors you use, some processors may have partitions that contain elements connected to the boundary, and others will have elements solely on the interior.
This bug is due to us putting the call to that extraction function in the wrong place: inside the boundary condition calculation for each type of boundary condition. As long as you use sufficiently few processors, every partition will contain at least some elements connected to the boundary, and this communication will work without any problem. However past a certain point, you start to create partitions that don't have any elements connected to the boundary. They will happily carry on through the communication and cause the process to deadlock.
Hope that explains the situation! I expect we'll be doing a patch release at the end of the month that will contain this fix.
Cheers,
Dave
On 11 Jul 2016, at 22:29, Selbach, Fabian <fabian.selbach@student.uni-siegen.de> wrote:
Hello,
sorry for my late response, but actually i tested a few things and still got some problems.
@Rodrigo: I tested the serial-run. It works on our Cluster and my own computer in the 3D-case! By the way is the bug-fix by David you mentioned, when you wrote how to fix the problem? Or do you have another idea how to fix this bug?
So it must be really a problem of parallelization due to openmpi in connection with Nektar++. After i tested the bug-fix (by David) on my computer, it runs in parallel-version! But actually i can't run it on our Cluster, because there is not installed an actually git-Version of Nektar++ with this bug-fix. So will there be a next Version of Nektar++ in the near future, which fixes this bug?
I even realized, that the parallel-run is possible in 3D by the Nektar++ 4.3.2 (without the bug-fix) on my own Netbook with an old version of openmpi (version 1.6), but why?
Best regards
Fabian ________________________________________ Von: nektar-users-bounces@imperial.ac.uk [nektar-users-bounces@imperial.ac.uk]" im Auftrag von "Selbach, Fabian [fabian.selbach@student.uni-siegen.de] Gesendet: Donnerstag, 7. Juli 2016 12:04 An: Mengaldo, Gianmarco; Moxey, David; Amitvikram Dutta Cc: nektar-users Betreff: Re: [Nektar-users] Problem 3D CompressibleFlowSolver
Hello,
thank you! Sure i will try it, but first i need to install a git repository version of Nektar++ on my computer. After that i will let you know about it.
@Gianmarco: the mesh size is about 1800000 points. The expansion is shown in the code below. @Rodrigo: i could not try a serial-run right now, but i will do.
Best regards
Fabian
________________________________________ Von: Mengaldo, Gianmarco [g.mengaldo11@imperial.ac.uk] Gesendet: Donnerstag, 7. Juli 2016 11:05 An: Moxey, David; Amitvikram Dutta Cc: Moura, Rodrigo C; Selbach, Fabian; nektar-users Betreff: Re: [Nektar-users] Problem 3D CompressibleFlowSolver
Hi Dave,
this sounds good, thanks!
As soon as Amitvikram and Fabian confirm that everything works fine, we can proceed to merge it.
Cheers,
Gian
Gianmarco Mengaldo
Department of Aeronautics and Mathematics
Imperial College London
SW7 2AZ, London, UK
*Currently at ECMWF
________________________________ From: David Moxey <d.moxey@imperial.ac.uk> Sent: 06 July 2016 21:36:59 To: Amitvikram Dutta Cc: Mengaldo, Gianmarco; Moura, Rodrigo C; Selbach, Fabian; nektar-users Subject: Re: [Nektar-users] Problem 3D CompressibleFlowSolver
Hi all,
Okay, I have now submitted a MR which should hopefully fix this issue:
https://gitlab.nektar.info/nektar/nektar/merge_requests/657
To test it you'll need to check out the source, which is in my local fork. You can check this out using the following commands:
git remote add dmoxey https://gitlab.nektar.info/dmoxey/nektar.git git fetch dmoxey git checkout -t dmoxey/fix/cfs-dgopt
Cheers,
Dave
On 6 Jul 2016, at 21:21, Amitvikram Dutta <amitvdutta23@gmail.com> wrote:
Sure!
On Wed, Jul 6, 2016, 12:55 PM David Moxey <d.moxey@imperial.ac.uk> wrote: Hi Gian,
In fact it seems I have a branch sat in my repository that fixes this but I apparently forgot to submit a merge request..!
I'll tidy it up and submit a request shortly, and then perhaps Amitvikram and Fabian, you can test this for us?
Cheers,
Dave
On 6 Jul 2016, at 20:51, Mengaldo, Gianmarco <g.mengaldo11@imperial.ac.uk> wrote:
Hi Dave,
Yep, indeed, this is the case! We should finally patch it out! ;)
Cheers, Gian
Sent from my Samsung device
-------- Original message -------- From: David Moxey <d.moxey@imperial.ac.uk> Date: 06/07/2016 20:32 (GMT+00:00) To: "Mengaldo, Gianmarco" <g.mengaldo11@imperial.ac.uk> Cc: Amitvikram Dutta <amitvdutta23@gmail.com>, "Moura, Rodrigo C" <r.moura13@imperial.ac.uk>, "Selbach, Fabian" <fabian.selbach@student.uni-siegen.de>, nektar-users <nektar-users@imperial.ac.uk> Subject: Re: [Nektar-users] Problem 3D CompressibleFlowSolver
Hi all,
I am guessing this is down to us using ExtractTracePhys() inside the boundary condition functions? I thought this had been patched out but it seems not.
Cheers,
Dave
On 6 Jul 2016, at 20:25, Mengaldo, Gianmarco <g.mengaldo11@imperial.ac.uk> wrote:
Hi guys,
Yes, this is a known issue that we are trying to fix permanently!
We have a temporary fix for it at the moment. Rodrigo should be able to help.
Regards, Gianmarco
Sent from my Samsung device
-------- Original message -------- From: Amitvikram Dutta <amitvdutta23@gmail.com> Date: 06/07/2016 19:22 (GMT+00:00) To: "Moura, Rodrigo C" <r.moura13@imperial.ac.uk>, "Selbach, Fabian" <fabian.selbach@student.uni-siegen.de> Cc: nektar-users <nektar-users@imperial.ac.uk> Subject: Re: [Nektar-users] Problem 3D CompressibleFlowSolver
Hi,
I've been having an identical problem and would appreciate knowledge of the bypass
On Wed, Jul 6, 2016, 11:20 AM Rodrigo Moura <r.moura13@imperial.ac.uk> wrote: Hello Fabian,
This might be an issue related to parallelization due to some recent modifications in the master or release branches. Does your case run in serial? If it runs, I can show you how to bypass this problem and run in parallel before an official fix is done.
Cheers, Rodrigo . _______________________________________________ Nektar-users mailing list Nektar-users@imperial.ac.uk https://mailman.ic.ac.uk/mailman/listinfo/nektar-users -- Amitvikram Dutta MASc Candidate Graduate Research Assistant Okanagan CFD Laboratory University of British Columbia | Okanagan Campus _______________________________________________ Nektar-users mailing list Nektar-users@imperial.ac.uk https://mailman.ic.ac.uk/mailman/listinfo/nektar-users
-- David Moxey (Research and Teaching Fellow) d.moxey@imperial.ac.uk | www.imperial.ac.uk/people/d.moxey<http://www.imperial.ac.uk/people/d.moxey>
Room 364, Department of Aeronautics, Imperial College London, London, SW7 2AZ, UK.
-- David Moxey (Research and Teaching Fellow) d.moxey@imperial.ac.uk | www.imperial.ac.uk/people/d.moxey<http://www.imperial.ac.uk/people/d.moxey>
Room 364, Department of Aeronautics, Imperial College London, London, SW7 2AZ, UK.
-- Amitvikram Dutta MASc Candidate Graduate Research Assistant Okanagan CFD Laboratory University of British Columbia | Okanagan Campus
-- David Moxey (Research and Teaching Fellow) d.moxey@imperial.ac.uk | www.imperial.ac.uk/people/d.moxey<http://www.imperial.ac.uk/people/d.moxey>
Room 364, Department of Aeronautics, Imperial College London, London, SW7 2AZ, UK.
_______________________________________________ Nektar-users mailing list Nektar-users@imperial.ac.uk https://mailman.ic.ac.uk/mailman/listinfo/nektar-users
-- David Moxey (Research and Teaching Fellow) d.moxey@imperial.ac.uk | www.imperial.ac.uk/people/d.moxey
Room 364, Department of Aeronautics, Imperial College London, London, SW7 2AZ, UK.
_______________________________________________ Nektar-users mailing list Nektar-users@imperial.ac.uk https://mailman.ic.ac.uk/mailman/listinfo/nektar-users
-- David Moxey (Research and Teaching Fellow) d.moxey@imperial.ac.uk | www.imperial.ac.uk/people/d.moxey Room 364, Department of Aeronautics, Imperial College London, London, SW7 2AZ, UK.