[Paraview] [EXTERNAL] Re: ... not yet supported for more than 2147483647 bytes.
Kashiwa, Bucky
bak at lanl.gov
Fri Jun 30 12:28:55 EDT 2017
Hey Alan: We originally encountered the Generic Warning in v4.3.1; Andy suggested trying 5.2 (see below). Subsequently (see below) we have tried v5.2, v5.3, and v5.4 - all of which behave much the same with large point numbers. I believe that Andy’s assessment is correct: this is an MPI issue. However, why ParaView should find it necessary to pass all of the point data, all at once, is kinda baffling. So maybe you have pointed to something to study: has the 4.* bug really been fixed? Especially in the case of point data (rather than grid data, which, in our case, is always considerably smaller). Thanks for writing, and thinking about the issue. Cheers, b.
=======================================================================73
<>Bucky Kashiwa PhD, PE <> Post: MS B216, Los Alamos, NM 87545 <>
<> Ofc: TA3-SM123-RM276 <>Email: bak at lanl.gov<mailto:bak at lanl.gov>, kashiwa at qwest.net<mailto:kashiwa at qwest.net> <>
<>Voice: 505-667-8812 <> Fax: 505-665-5926 <>
<> Home: 505-988-7332 <> Cell: 505-795-5581 <>
=======================================================================73
From: "Scott, W Alan" <wascott at sandia.gov<mailto:wascott at sandia.gov>>
Date: Thursday, June 29, 2017 at 11:31 AM
To: Bucky Kashiwa <bak at lanl.gov<mailto:bak at lanl.gov>>, "Bauer, Andy (External Contacts)" <andy.bauer at kitware.com<mailto:andy.bauer at kitware.com>>, andrealphus <andrealphus at gmail.com<mailto:andrealphus at gmail.com>>
Cc: "ParaView at ParaView.org<mailto:ParaView at ParaView.org>" <ParaView at paraview.org<mailto:ParaView at paraview.org>>
Subject: RE: [EXTERNAL] Re: [Paraview] ... not yet supported for more than 2147483647 bytes.
I notice you are running ParaView 4.3.1? We had a bug in the earlier 4.*.* versions where we were passing massive amounts of information from all processes to all processes on the state of the status bar, bottom of the screen. We stopped doing that. I wonder if this could be an issue with huge numbers of points?
Try PV 5.4.0? (Or possibly wait for 5.4.1, out Real Soon Now?)
Alan
From: ParaView [mailto:paraview-bounces at paraview.org] On Behalf Of Kashiwa, Bucky
Sent: Thursday, June 29, 2017 10:53 AM
To: Bauer, Andy (External Contacts) <andy.bauer at kitware.com<mailto:andy.bauer at kitware.com>>; andrealphus <andrealphus at gmail.com<mailto:andrealphus at gmail.com>>
Cc: ParaView at ParaView.org<mailto:ParaView at ParaView.org>
Subject: [EXTERNAL] Re: [Paraview] ... not yet supported for more than 2147483647 bytes.
Andy, Ashton: We have now tried versions 5.2, 5.3, and 5.4. With large point data sets we still have the same Generic Warning cited below (followed by ERRORS that cause PV to hang). I reckon that we are hitting the MPI wall associated with 2^31 items communicated at once - as per Andy’s note below.
All of our recent work has been using the downloadable binaries from paraview.org, for both the client and the server. We see the same behavior using the MacOS client to Linux server, and Linux client to Linux server.
This MPI shortcoming is a serious showstopper for us so we are going to have to find a remedy.
The first question is: why should either the client or the server think that more than 2^31 items need to be communicated in the first place? On the surface, this seems to be unreasonable.
To be clear, there is no problem opening and reading the datafiles, and displaying the outline view. The warning appears when we try to display the point view, when the number of points is too large. If we take a slice that reduces the number of points to below some threshold, then the point view display is okay. As an example, consider a .cosmo64 formatted point data file. There are 36 Bytes of data per point. We can display ~28,000,000 points and ~35,500,000 points just fine. At 45,500,000 points (=1.638 GBytes) the Generic Warning gets thrown. This seems to be independent of the number of nodes and PEs used by the server, and whether the view RenderView or EyeDomeLighting. We also write/read .vtm format that behaves in a similar fashion: slices will display okay until the slice is thick enough that there are too many points in the image.
I have tried doing MPI on the client, which seems to have no effect on the foregoing limitation on the number of displayable points. Please let me know if you can think of other switches that can be thrown, that may shed some more light on the issue, or if you are aware of a forthcoming repair. Thanks very much, b.
=======================================================================73
<>Bucky Kashiwa PhD, PE <> Post: MS B216, Los Alamos, NM 87545 <>
<> Ofc: TA3-SM123-RM276 <>Email: bak at lanl.gov<mailto:bak at lanl.gov>, kashiwa at qwest.net<mailto:kashiwa at qwest.net> <>
<>Voice: 505-667-8812 <> Fax: 505-665-5926 <>
<> Home: 505-988-7332 <> Cell: 505-795-5581 <>
=======================================================================73
From: Andy Bauer <andy.bauer at kitware.com<mailto:andy.bauer at kitware.com>>
Date: Monday, December 19, 2016 at 4:17 PM
To: andrealphus <andrealphus at gmail.com<mailto:andrealphus at gmail.com>>
Cc: Bucky Kashiwa <bak at lanl.gov<mailto:bak at lanl.gov>>, "ParaView at ParaView.org<mailto:ParaView at ParaView.org>" <ParaView at paraview.org<mailto:ParaView at paraview.org>>
Subject: Re: [Paraview] ... not yet supported for more than 2147483647 bytes.
There are two parts to this issue. The first is that that vtkMPICommunicator for PV 4.3.1 won't communicate data that is over 2^31 bytes of data. This is fixed in PV 5.2. The other issue is due to MPI having a limit of 2^31 objects to be communicated in a single shot. This is MPI's API in that the count for objects that are typically sent/received is an int. See http://www.mpich.org/static/docs/v3.1/www3/MPI_Send.html for example.
On Mon, Dec 19, 2016 at 4:59 PM, andrealphus <andrealphus at gmail.com<mailto:andrealphus at gmail.com>> wrote:
That is a 32 bit error, from trying to index something with more than
(2^32)/2 elements or indices. Are you using any custom
libraries/packages/modules which might not be 64 bit compliant? Are
you sure you built a 64 bit version (check your gcc -v).
-ashton
On Mon, Dec 19, 2016 at 1:32 PM, Kashiwa, Bucky <bak at lanl.gov<mailto:bak at lanl.gov>> wrote:
> On Linux using ParaView version 4.3.1 built with OSMesa-9.0.1, OpenMPI,
> etc. Running pvserver with 12 PEs, client-server mode, Standard Release
> ParaView-4.3.1-Linux-64bit client. With large point data (>2Gb) we get
> this error, upon trying to display a large number of points:
>
> Generic Warning: In
> ../ParaView-v4.3.1-source/VTK/Parallel/MPI/vtkMPICommunicator.cxx,
> line 194
>
> This operation not yet supported for more than 2147483647 bytes.
>
> Our CMakeCache.txt is attached, in case it may provide helpful clues.
>
> Thanks much. B. Kashiwa
>
>
>
> _______________________________________________
> Powered by www.kitware.com<http://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
>
> Search the list archives at: http://markmail.org/search/?q=ParaView
>
> Follow this link to subscribe/unsubscribe:
> http://public.kitware.com/mailman/listinfo/paraview
>
_______________________________________________
Powered by www.kitware.com<http://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
Search the list archives at: http://markmail.org/search/?q=ParaView
Follow this link to subscribe/unsubscribe:
http://public.kitware.com/mailman/listinfo/paraview
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/paraview/attachments/20170630/b88ccbe2/attachment.html>
More information about the ParaView
mailing list