[Paraview] Connecting to pvserver

[David E DeMarle]

On Thu, Jan 17, 2013 at 2:12 AM, Pradeep Jha
<pradeep.kumar.jha at gmail.com> wrote:
> Hello,
>
> I am trying to use PV for the first time in parallel. I have some basic
> questions.
>
> I have installed PV 3.98 from source on my local machine (Mac OSX). I don't
> have the root permissions for the remote system I want to use for

You shouldn't need root permissions. You can build paraview in your
home directory (or anywhere that the remote machine can read from) and
connect to that. You generally don't even have to run make install.

> visaulization. So I installed the binary form of PV 3.98 on the remote

Our binaries typically don't have an MPI in them that will work on the
cluster's network. What we have is more or less just for making use of
the cores on a single node. So you typically have to compile paraview
on the cluster and enable MPI to use it in distributed memory parallel
settings. It should work for you on the remote node, but not the
cluster.

> system. I set up the server on that machine using the command:
>
> $mpiexec -n 4 pvserver --use-offscreen-rendering
>
> And then I am trying to connect to this server from the PV on my Mac. But it
> is giving me the following error:
> -----------------------------------------------------------------------------------------------------------------------------------------/Users/pradeep/softwares/ParaView/ParaViewCore/ClientServerCore/Core/vtkTCPNetworkAccessManager.cxx,
> line 332
> vtkTCPNetworkAccessManager (0x7f953b63aa30): Failed to connect to
> localhost:11150. Client-Server Handshake failed. Please verify that the
> client and server versions are compatible with each other
>
> Any ideas what is going wrong?

This is most often happens when the client and server are mismatching
versions (3.14 connecting to 3.10 etc), but various network problems
can result in the same thing.

>
> Also, this remote system has 16 processors. But for bigger jobs, I can
> submit a job on a cluster using this remote system using a shell script
> where it can use over 100 processors. But I cannot install anything on this
> cluster. ...

Same comment as before, you only have to have a paraview built
somewhere that the compute nodes can access, so you shouldn't have to
install it.

> ... Can I use this remote cluster for using paraview faster?
>

A word of caution - more nodes does not necessarily mean faster.
ParaView's parallel architecture allows it to process larger data and
larger data as more disks and memory are added - not the same sized
data faster and faster using more CPUs and GPUs are added (accept in
the infrequent case when processing and rendering are the
bottlenecks).

> The structure is something like this: Mac ---> Remote system (16 processors)
> ----> Cluster with over 100 nodes.

You'll need to set up an ssh tunnel through the remote (login) node to
the cluster in your job submission to do that.

> Thanks,
> Pradeep
>


David E DeMarle
Kitware, Inc.
R&D Engineer
21 Corporate Drive
Clifton Park, NY 12065-8662
Phone: 518-881-4909