Instructions for using the visualization server wzero
Connecting to wzero
The machine called wzero is the dedicated visualization machine among the quarks. Unlike the other machines, it has a graphics card that greatly accelerates visualization.
To use wzero, you can just do ssh wzero but because of the way the X Window System works that is used to display windows on your screen for programs that run on a different machine, doing just ssh will do you no good: this simply transfers all information that needs to be rendered across the network and will then be rendered on your local machine. This is, then, clearly not the way to go to render large data sets.
Rather, you need to connect to wzero in a way that allows all rendering operations to happen on that machine. This can be achieved by logging into wzero using the command
vglconnect -s wzero
(vgl stands for
"virtual graphics
library" and is based on
the OpenGL
libraries; the -s
flag transports the graphical output
via ssh). Then, call whatever program you want to start using
the vglrun command, for example as in
vglrun visit
to call the Visit visualization program.
Parallel visualization with Visit
The Visit visualization program can use multiple processors to visualize data, greatly accelerating the process for large data sets. wzero has 32 cores that are available for this and they can be used by calling Visit as follows:
vglrun visit -par -np 32
Parallel visualization with Paraview
Setup the server
### In a Terminal Window local> ssh wzero; mpirun -np 32 pvserver --server-port=xxxx (choose a port number, must be different for each user
Setup the client
### In a Different Terminal Window (added a -4 below for ipv4 now) local> ssh -4 -L 11111:localhost:xxxx wzero -f sleep 3600 (xxxx is the same port number from above, sleep 3600 means it lasts one hour) local> paraview In Paraview Edit->Settings->Server uncheck Remote Render Threshold File->Connect (You have to configure it the first time) Add Server Name Click configure/save Connect
At this point, you are running a paraview server on the quark and forwarding it back to a paraview client running on your office machine