[Paraview] ParaView pvserver crashes

[Moreland, Kenneth]

Jay,

Have you compiled with OSMesa support (an additional step to compiling with Mesa) and run with -use-offscreen-rendering flag (I can't check your script because my email client stripped it) as is described in the ParaView Wiki (http://www.paraview.org/Wiki/Setting_up_a_ParaView_Server#OSMesa_support)?  It could be that pvserver is still trying to open an X connection, failing, and crashing.

Another possibility is that there are multiple OpenGL libraries on your system and multiple ones are being grabbed.  Unfortunately, this can be pretty tricky to narrow down if that is the case.  You might try building your own version of Mesa from source as you suggest.  Mesa is actually pretty easy to build, and it should not add any complexity to building ParaView, which you have to do anyway if you plan to run on a server with more than one process.

-Ken


On 5/27/09 9:25 PM, "Jay Oswald" <jjo.lists at gmail.com> wrote:

Hello,

I tried setting up a ParaView server on a university cluster, but am getting the server to crash when remote rendering is enabled.  First off, I suppose I should ask if I am wise to do this.  I don't see any rules of thumb for how large a dataset should be for good scaling, or how fast an interconnect ParaView requires.

I generate fairly large unstructured datasets (in the range of 1-3 million nodes).  My data files are in the exodus format and generally run in the hundreds of MB.  My connection to the cluster is limited at 100 MBit, and the interconnect between cluster nodes is Gigabit.  Even if the speed is comparable, its nice enough not to need to download the datafiles to my computer.

The cluster is made up of dual socket quad core machines with pitiful on-board graphics.  I built ParaView 3.4 on this system with OpenMPI and Mesa support.

I run the server through a PBS scheduler.  Because the compute nodes are not directly connected to the outside world, and because I don't know apriori which nodes the ParaView server will run on, I use a reverse connection.  My PBS script is attached (paraview_server.sh).  It connects fine, and I am able to open a dataset.  Once I do something that requires a fair amount of work (i.e. switching from outline view to surface view) the server crashes.  I have also attached the output from the server captured by PBS (o_paraview.log).  Finally so that you can see what I've configured, I've attached the CMakeCache.txt file from my ParaView build.  Note that in my script, I've commented out the remote rendering option, it does not crash if this is left commented out.

Other relevant data is that the cluster is running Ubuntu 8.04 x86_64, and I'm running the 64bit version of ParaView on Windows Server 2008.  I'm using the Mesa library provided by the Ubuntu repositories (7.0.3).  The crash also happens identically if I run the ParaView client over a remote X session directly from the cluster, so I'm certain its not a problem with mismatched versions.

My ideas at this point are:
- try building Mesa from source (I'd rather not muck around with having the repository + custom built version however)
- live without remote rendering (which is quite slow)
- hope someone on the ParaView list has a better idea

Any suggestions would be greatly appreciated.

Jay Oswald
Northwestern University






   ****      Kenneth Moreland
    ***      Sandia National Laboratories
***********
*** *** ***  email: kmorel at sandia.gov
**  ***  **  phone: (505) 844-8919
    ***      web:   http://www.cs.unm.edu/~kmorel

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.paraview.org/pipermail/paraview/attachments/20090623/5d26cb11/attachment-0001.htm>