ParaView¶
Paraview is supported software on Alps. See the main applications page for more information.
ParaView is an open-source, multi-platform scientific data analysis and visualization tool, which enables analysis and visualization of extremely large datasets.
ParaView is both:
- a general purpose end-user application with a distributed architecture, that can be seamlessly leveraged by your desktop or other remote parallel computing resources, and
- an extensible framework with a collection of tools and libraries for various applications, including scripting (using Python), web visualization (through Trame and ParaViewWeb), and in situ analysis (with Catalyst).
ParaView is provided on ALPS via uenv.
Before Starting¶
The very first step is pulling a ParaView uenv (see uenv quick-start guide on how to do it). Pick the latest version (or the one you prefer) from the list of available uenvs given by
then pull it with something like
and you’re ready to start.
The guide assumes that user has just a single ParaView uenv image in their local repository.
Over time you might end up with multiple ParaView uenvs pulled in your local repository.
But for the sake of simplicity, this guide assumes that uenv image ls paraview list one and exactly one uenv.
If it lists more, you will have to use a more specific uenv label in all code/file snippets in this guide when there is a reference to the uenv to disambiguate which one to use.
One-time setup¶
CSCS provides helper scripts that are very handy for launching live sessions and batch rendering.
Warning
If you use just a single uenv image on a single system, the following setup works nicely. You just have to remember to update the scripts accordingly, if you switch to a more up-to-date uenv image.
But you might end up with problems if:
- you use different uenv images on the same system
- you use the same uenv image version but on different systems (e.g. paraview/6.0.1:v1 on gh200 and zen2)
Mainly because helper scripts might not be cross-platform nor cross-version compatible. Have a look at how to manage different platforms configuration for what concerns the cross-platform compatibility, while for what concerns cross-version compatibility you should consider in your setup also the uenv labels.
To install these utilities, the simplest approach is to place them in a directory that is part of your PATH.
A common convention is to create a personal ~/bin directory and add it to your PATH.
mkdir ~/bin && echo 'export PATH=~/bin:$PATH' >> ~/.bashrc && source ~/.bashrc
uenv run paraview -- cp -r /user-environment/helpers/. ~/bin
You can then test that helpers scripts are installed with
$ paraview-reverse-connect
Usage: paraview-reverse-connect <uenv-label> <server-port> [<srun-option>]*
Different uenvs might contain different scripts
You can check what helper scripts a ParaView uenv provides with
Whenever you switch to a different uenv image, it is advisable to reinstall also the scripts by using the one provided by the new uenv image.
Running ParaView in batch mode with Python scripts¶
The following sbatch script can be used as template for running ParaView in batch mode.
Note
Current observation is that best performance is achieved using one MPI rank per GPU. How to run multiple ranks per GPU is described here.
Using ParaView in client-server mode¶
Make sure to use the same version of ParaView on both sides.
A ParaView server can connect to a remote ParaView client installed on your workstation.
To do that, your local ParaView client needs to connect to a pvserver running on Alps compute nodes, which is started using a SLURM job with appropriate parameters.
It can be done manually each time, or ParaView can be configured to do that for you auto-magically. 🪄
Connecting using an PVSC configuration file¶
A ParaView Server Configuration (PVSC) file is an XML file that contains one or more server configurations.
This is a very simple configuration that at each connection to Alps Daint it prompts you for:
- uenv image label to use (it will be stored for next time)
- SLURM arguments for the allocation (every time it proposes the default value)
- TCP port to use for the reverse connection
<Servers>
<Server name="CSCS Alps Daint" resource="csrc://daint.cscs.ch:11111" timeout="-1">
<CommandStartup>
<Options>
<Option name="UENV" label="ParaView uenv" save="true" readonly="false">
<String default="paraview"/>
</Option>
<Option name="SLURM_ARGS" label="SLURM Arguments" readonly="false">
<String default="-n4 -pdebug -t10"/>
</Option>
<Option name="PV_SERVER_PORT" label="Server Port" readonly="false">
<Range type="int" min="11111" max="65535"/>
</Option>
</Options>
<SSHCommand exec="paraview-reverse-connect" delay="0" process_wait="0">
<SSHConfig>
<PortForwarding/>
</SSHConfig>
<Arguments>
<Argument value="$UENV$"/>
<Argument value="$PV_SERVER_PORT$"/>
<Argument value="$SLURM_ARGS$"/>
</Arguments>
</SSHCommand>
</CommandStartup>
</Server>
</Servers>
Tip
You can use this as a starting point for more advanced setups and customizations.
See the official documentation about PVSC files for examples and details.
An easy way to add a server configuration to your ParaView is to create it in a file with *.pvsc extension, and then load it in ParaView with File → Connect… → Load Servers.
This will add the configuration(s) provided to the list of available server proposed when you click on File → Connect….
You can manipulate the list of server configurations from the dialog, and changes will be reflected (on ParaView UI exit) in a file called servers.pvsc in your ParaView user settings directory.
A more advanced and versatile configuration¶
This is the most versatile way to configure ParaView client-server connection to Alps, as it allows you to customize (any part of) the command directly from the ParaView UI.
You can achieve the same exact results either by creating a PVSC file as described in the previous section with the following content:
<Servers>
<Server name="CSCS Alps" resource="csrc://localhost:10111" timeout="-1">
<CommandStartup>
<Command process_wait="0" delay="0" exec="ssh">
<Arguments>
<Argument value="-R $PV_SERVER_PORT$:localhost:$PV_SERVER_PORT$"/>
<Argument value="daint.cscs.ch"/>
<Argument value="--"/>
<Argument value="paraview-reverse-connect"/>
<Argument value="paraview"/>
<Argument value="$PV_SERVER_PORT$"/>
<Argument value="-n4 -pdebug --gpus-per-task=1"/>
</Arguments>
</Command>
</CommandStartup>
</Server>
</Servers>
Or by adding a new server configuration directly from the ParaView UI:
- File → Connect… → Add Server
- Specify a name for the configuration
- Select Reverse Connection
- Click on Configure
- Select Startup Type: Command
And type in the following command:
ssh -R $PV_SERVER_PORT$:localhost:$PV_SERVER_PORT$ daint.cscs.ch -- paraview-reverse-connect paraview $PV_SERVER_PORT$ -n4 -pdebug --gpus-per-task=1
Understanding and customizing the command¶
Let’s dissect the command in order to fully understand it, so you can customize it for your needs.
In the command it is possible to identify two parts separated by “--”:
- SSH connection
- ParaView server launch with
paraview-reverse-connect
SSH connection¶
This first part runs locally on your workstation and specifies how to connect to Alps via SSH.
You should use whatever SSH option you are normally using to connect to Alps.
What’s important is having -R $PV_SERVER_PORT$:localhost:$PV_SERVER_PORT$, which is responsible of forwarding the port specified in the GUI (if it is busy, you can try a different one) from your local workstation to Alps.
ParaView server launch with paraview-reverse-connect¶
The second part uses paraview-reverse-connect (see how to obtain it), which runs on the Alps login node to launch a SLURM job.
It will execute ParaView pvserver instances on compute nodes, that will (reverse) connect with your ParaView UI on your workstation.
The first two arguments are required, and they are:
- the uenv image label and
- the port you are forwarding via SSH.
After them, it is possible to specify any srun option, giving you full control on the allocation request (e.g. time, partition).