[Paraview] Run Paraview on SVA

Weiguang Guan guanw at rhpcs.mcmaster.ca
Thu Nov 29 17:06:51 EST 2007


Thanks so much for your reply. Below are my comments...

On Thu, 29 Nov 2007, Moreland, Kenneth wrote:

>>
>> (1) When using reverse connect, it prints out the following error:
>>
>> ICET,0:ERROR: Sizes of src and dest do not agree.
>>
>> every time Paraview refreshes its rendering.
>
> That error usually occurs when the available display on node 0 is larger
> than one or more of the other displays in your cluster.  I'm surprised
> this only happens in reverse connect, it generally has nothing to do
> with the direction of the connection.  At any rate, work with your
> system administrators to make sure X Windows is set up to have the same
> size display on all nodes in your cluster.
>

I think you are right. It's likely that I did the forward and reverse 
connection tests, one before and the other after our SVA was reconfigured.

>>
>> (3) Except for the first render node, all the the other render nodes
>> consume 100% CPU constantly.
>
> That is not uncommon with some implementations of MPI.  When the server
> is sitting at idle, node 0 is listening for a socket while all others
> are sitting in an MPI_Recv call.  Some implementations of MPI must
> implement MPI_Recv in a loop that constantly polls for incoming data (I
> guess).  At any rate there is nothing we can really do about it.
>
> -Ken
>
Does this affect the performance of paraview? Can we consider it as a bug 
of MPI lib (we use HP MPI lib)?

Weiguang


More information about the ParaView mailing list