[Paraview] signal 11

[Dave Semeraro]

Hi Berk,

I have more info on the signal 11 problem. I have done some more experiments.
First of all I can read the dem file using paraview on my workstation. The 
workstation
arch is the same as the cluster node arch. That is 32 bit dual Xeon running 
Fedora
core 2.0. Reading the file with paraview on my windows box ( Dell laptop with
windows xp and 2gig of ram etc ) crashes paraview. Reading the file on the 
cluster
with pvserver running on a single node and pvclient running in my office 
crashes
both pvserver and pvclient. The pvserver does not signal 11 on the node it 
just exits
and you get a message "All remote MPI processes have exited" from the verbose
version of mpirun. It segfaults the pvclient. One other interesting 
behavior is that when
I read the file on my workstation with paraview I get an error after I 
close paraview.
The error is:
glibc detected *** free(): invalid pointer: 0xf18dd008



If I convert the .dem file to a .pvd file on my workstation things are 
different. All the
above tests work. I can read the file on the workstation and laptop with 
paraview and
I can read the file with the cluster running pvserver and my workstation as 
the
pvclient. No errors no problems.

I got the dem file from a guy down the hall. He
does a lot of GIS work and monkeys with the dem files a lot. He told me 
that a true
USGS dem file is supposed to be square. It is supposed to have the same 
number of
points in each direction. The file he gave me is not square. I dont know if 
the dem files
are really supposed to be square or not. But that would help to explain the 
reason
why the dem reader might not work quite right if it expected a square file 
and got
a rectangular one instead. It might not be mallocing enough memory and seg 
faulting
once it tries to write the rectangular data into the square space. I will 
try to get a
square dem of roughly the same size as the one that fails and see if that makes
a difference.

Dave
At 10:27 AM 6/27/2006, Berk Geveci wrote:
>Hi Dave,
>
>Have you tried with fewer nodes? 2 for example. You should be able to 
>attach a debugger to each node if you can reproduce the problem in a more 
>controlled environment. Also, have you tried converting the file to 
>another format (VTK for example) and reading it that way?
>
>-Berk
>
>
>On 6/26/06, Dave Semeraro 
><<mailto:semeraro at ncsa.uiuc.edu>semeraro at ncsa.uiuc.edu> wrote:
>
>Hello,
>
>I am running paraview 2.4 on a linux cluster ( 32 bit, myrinet connectivity
>). I am
>having problems reading a dem file on the cluster. The file is about
>30Mbytes in
>size so it isnt overly large. I can read it on a a single paraview instance
>on my
>local workstation. When I try to read it on the cluster using 4 nodes I get  a
>series of errors that look like this:
>
>1: signal 11 received, exiting..
>1: Signal sent from unknown source.
>
>I get this message from instance 1, 2, and 3 and paraview dies on those 
>nodes.
>Has anyone else seen this kind of behavior? I dont know if this is an MPI
>thing or
>a paraview thing. Any ideas on how to trace the cause of this error down? I
>know
>these questions are rather vague and I am sorry for that. I have tested
>this cluster
>set up on a small ( 50^3) hdf dataset and things behave fine.  Since I can
>read the
>dem data on a single node I assume the reader isnt the problem. So any 
>pointers
>on where to look next would be a big help. Thanks.
>
>Regards,
>Dave Semeraro
>
>
>_______________________________________________
>ParaView mailing list
><mailto:ParaView at paraview.org>ParaView at paraview.org
>http://www.paraview.org/mailman/listinfo/paraview
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://public.kitware.com/pipermail/paraview/attachments/20060627/9e6ce3db/attachment.html