[Sloy]

Hello Community,

At the moment I am evaluating to buy a post processing workstation for RANS cases of up to (not always) 40-50 million cells. At the moment I have two options, buying one workstation with a 4820K and 64GB of ram or buying two workstations with a 4820K and 32GB each.

I did some small research and saw that it is possible to run Paraview in server/client mode. But it appears to be that the server and the client cannot be the same computer. So is it possible to run paraview in parallel on two workstations at the same time where one workstation is just one "server node" and the second workstation is a "sever node" and a "client"?

Is this sever/client mode only working for GPU rendering or is it also possible to spread the case across the two memories of workstations?

Regards

Sloy

[wyldckat]

Greetings Sloy and welcome to the forum!

ParaView is able to do whatever you need to do, in the sense that you can place the server applications running wherever you want, as long as all of them have access to the necessary data on disk or shared folder. For a lot of information on this topic, read this page: http://paraview.org/Wiki/Setting_up_a_ParaView_Server

In summary: You can easily run the server applications (to act like a single server) like this:

mpirun -np 8 --hostfile machines.txt ./pvserver

Where the file "machines.txt" would define the list of machines and number of cores to assign to each machine.

As for the client application, it can be wherever you want it to be. It's only that the client-server structure that was implemented in ParaView was developed for having the client in a separate machine. This is likelly related to this: you might only want 1 server application per machine (where the 2 would be working together), depending on the data to be open and in order to at least reserve some CPU power for the client application in the main machine. Otherwise, you'll have the client and the server fighting for CPU power on the main machine.

The downside of this strategy is that you must open data that is already decomposed and/or with a parallel-ready reader, so that you can take advantage of the 32GB on each machine. Otherwise, you'll still be limited to only 32GB in total.

Best regards,
Bruno