Using remote desktops on Casper¶

Using VNC¶

Most programs on Casper and Derecho are designed to run in the terminal, but a few either require the use of a graphical interface or work best in one. While using the default method – X-forwarding – is sufficient for simple programs like text editors, it can be prohibitively slow for more complex programs like MATLAB, IDL, VAPOR, and RStudio.

When you log on to your workstation, you typically interact with programs using a graphical desktop shell. With virtual network computing (VNC), you can use a graphical desktop shell to work remotely on the Casper data analysis and visualization cluster.

The remote desktop runs on Casper in a VNC session that is managed by the VNC server. You access it with a VNC client, which runs on your local workstation.

To get started¶

Download and install a VNC client on your local machine. CISL recommends the TigerVNC client and provides this video to help Mac users install it: Installing TigerVNC on a Mac laptop. (Installing on Windows machines is less complex.)

TurboVNC also works, but systems using Java versions >8 do not include the necessary runtime libraries to enable TurboVNC's VNC viewer. Some other VNC clients – RealVNC, for example – do not work well with the VNC software installed on Casper.

Connecting to a VNC session¶

To begin using a remote desktop with VNC, you will need to start a VNC session. Your session will run within a Casper job and can persist for up to 24 hours.

These basic steps for starting a session are described in detail below:

• Run the vncmgr script, which enables you to configure your VNC desktop session and start both a Casper batch job and the VNC server.

• Connect to the Casper batch node with your VNC client, using a port that VNC specifies.

• Enter a one-time password to access your VNC session.

If you are connected to the NCAR Internal Network or using the NCAR VPN, you will be able to connect directly to your session on the Casper batch node using the VNC client.

If you are not on the NCAR network or VPN, you will need to create an SSH tunnel to connect your local machine and remote desktop. How to create the SSH tunnel is described below.

Overview of the vncmgr script¶

CISL provides the vncmgr script for initiating and managing VNC sessions on Casper. How to run it is described in detail in the following section. It can be used in interactive mode or command-line mode.

Interactive mode¶

If you run the script without any command-line arguments, it will launch in the interactive mode. In this mode, you can start a new session, list existing sessions, query a session to retrieve connection instructions and obtain a new one-time-password, and kill a running session. The script enables you to name your session, state how long you want the server to run, and choose which GPU you'd like to request for rendering support (GP100 or L40). It also allows for custom requests to both the job scheduler and the VNC server program.

Command-line mode¶

In command-line mode, you specify a subcommand and provide any desired options as command line arguments. Here are the available commands:

vncmgr list

vncmgr create [SESSION] --account PROJECT [--time WALLTIME …]

vncmgr query [SESSION]

vncmgr kill [SESSION]

Choosing a name SESSION is optional. If you do not provide a name, the name default will be assigned and referenced in each subcommand.

Customizing the Casper job and VNC server¶

The vncmgr script allows you to customize both the Casper session in which the server will run and the server itself. This customization can be done in both interactive and command-line modes. The most common uses involve increasing the resources allocated to your job. For example, you could allocate 4 CPUs and 20 GB of memory to a 2-hour VNC session using the command-line mode as follows:

vncmgr create \
       --account PROJECT \
       --time 2:00:00 \
       --job-opts="-l select=1:ncpus=4:mem=20GB"

vncmgr create \
       --account PROJECT \
       --time 2:00:00 \
       --job-opts="-l select=1:ncpus=4:mem=20GB" \
       --gpu-model=l40

vncmgr create \
       --account PROJECT \
       --time 2:00:00 \
       --job-opts="-l select=1:ncpus=4:mem=20GB -l gpu_type=l40"

By default, all VNC jobs are automatically placed on nodes with NVIDIA Quadro GP100 GPUs. You can optionally request that NVIDIA L40 GPUs are used instead either via the --gpu-model flag to vncmgr or by specifying -l gpu_type=l40 in your --job-opts. The L40 GPUs are faster than the GP100s and provide significantly more video RAM, so they are a good choice for more resource-intensive rendering tasks.

Run vncmgr --help in a Casper login session for more information about using the script and customizing your session.

Running vncmgr from your local machine¶

You can run the vncmgr command directly from your local machine as shown in the example below without first starting a login session on Casper. While both command-line and interactive mode will work, CISL recommends interactive mode as it will allow you to generate new one-time passwords via the query option without having to authenticate to Caspe every time.

Example VNC Session¶

This demonstrates how to run vncmgr, create and configure a customized VNC session, and then connect to the session with a VNC client. In this example, the user is not connected to the NCAR VPN and needs to use an SSH tunnel. (Alternative for PuTTY users.

1. Run this command to get started, using your own username:

   ssh -t -l username casper.hpc.ucar.edu /glade/u/apps/opt/vncmgr/bin/vncmgr
   

   You will be prompted to authenticate, after which you will have access to the vncmgr menu. (If you exit vncmgr, you will have to rerun the command to regain access to it.)

2. Create your new VNC session with a name up to 10 characters long. The session is named "vapor" in this example. (Choose names that will help you avoid confusion if you run multiple sessions.)

   

3. A series of prompts on the next screen (below) will ask you to specify:

   • Your project code.
   • The wallclock time you want in HH:MM:SS format. The default is 4 hours and the maximum is 24 hours.
   • Whether you wish to render the desktop itself on the GPU using VirtualGL. In general, we recommend leaving this off and using vglrun to launch applications which need GPU acceleration. Some applications (e.g., IDL) may become unstable with GPU acceleration active in the desktop.
   • Which GPU you wish to use for hardware acceleration - a GP100 (default) or an L40 GPU.
   • Any optional arguments. In the example, the user requests 20 GB of memory.

   All VNC jobs must run on a node with a visualization GPU. If you specify custom resource requirements, those requirements will be modified if necessary to ensure that the job can run on the correct node.

   

4. When the job starts, follow the instructions and choose how you want to connect to your VNC session: tunneling or using the UCAR internal network or VPN.

   

   Pressing enter at this point will return you to the interactive menu. Use the query option if you need to return to the instructions for creating the SSH tunnel later.

5. Your desktop on Casper will be displayed after you connect to the specified host and enter the onetime password. On the desktop, start a Konsole (terminal) from the list of applications in the launcher.

   

6. To start an application, load any required modules and run the executable.

   

   Here, we have used the vglrun command to execute the program VAPOR with OpenGL acceleration on the GPU. If you do not use vglrun, you will run your program with software rendering on the CPU. For more information on running VAPOR, see this documentation.

SSH tunneling with PuTTY¶

The general process for creating an ssh tunnel described above will work well for clients with a command line ssh installation, however additional steps are required for Windows users with PuTTY as their ssh client. Expand the note below for additional details.

ssh -l username -L xxxx:localhost4:xxxx casperxx.hpc.ucar.edu "bash .vnctunnel-default"

Using FastX¶

The FastX remote desktop service gives users access to Casper for performing lightweight tasks such as text editing, running programs such as xxdiff and ncview, or running analysis scripts that consume little in the way of graphics resources.

A user can log out of a FastX remote desktop and return to it later. This service will remain available while NCAR and UCAR building closures are in effect.

FastX sessions that consume excessive resources are subject to being killed. For resource-intensive workloads that do not require GPU-based rendering, consider starting a Casper job instead by running a qinteractive command from a FastX terminal window. For more resource-intensive work with high-end, GPU-accelerated graphics, consider using VNC as described above rather than FastX.

FastX can be accessed through a web browser or a desktop client. How to use both of these options is described below.

FastX via web browser with VPN access¶

To use FastX without installing any software, connect to the NCAR VPN and use an updated version of any common browser. See the following section for an alternative to using the VPN.

1. Connect to the NCAR VPN.

2. Go to https://fastx.ucar.edu:3300.

3. Authenticate with your username and token response.

4. Click the + button in the upper-left corner of the FastX window.

   

5. In the next window, click the KDE button and then Launch.

   

6. The KDE desktop will open in a new tab or a new browser window. Right-click on the desktop to start a Kconsole terminal window.

   

# source bash/zsh login environment definition scripts:
source /etc/profile

# source tcsh login environment definition scripts:
source /etc/csh.login

Reconnecting to your FastX session¶

If you want to retain your session to return to it later, just log out or kill the browser window. When you log in again, select the icon to re-open the session.

Terminating your FastX session¶

If you do not want to retain your session, terminate it as shown here before logging out:

FastX via web browser and ssh tunnel¶

If you are not connected to or do not want to connect to the NCAR VPN, you can still use FastX via web browser by creating an ssh tunnel from your laptop or desktop to FastX.

1. Start by running the following on your command line, inserting your own username.

   ssh -L 3300:fastx.ucar.edu:3300 username@fastx.ucar.edu
   

2. Authenticate as usual and you will be in a terminal session on a Casper node.
3. Leave that terminal session running, open your browser, and go to https://localhost:3300/. You may see a warning about the site being unsafe, but you may ignore the warning and continue.
4. Authenticate with your username and token response.
5. Continue as described in the previous section to launch the KDE desktop.

Alternatives for creating ssh tunnel¶

To create a tunnel using PuTTY or SecureCRT, following the examples in these video demonstrations:

• PuTTY ssh tunnel
• SecureCRT ssh tunnel

After creating the tunnel, proceed as described in the previous section.

FastX via desktop client¶

If you’re not connected to the NCAR VPN or prefer to use a faster, more robust remote desktop service, consider downloading and installing the FastX desktop client.

1. Start the FastX client.

2. Click the + button in the upper-left corner.
   

3. Fill in the fields of the pop-up boxes as follows:

   • Host: fastx.ucar.edu
   • User: Enter your username
   • Port: 22
   • Name: CISL remote desktop (or any string)
   • Run (advanced tab): /ncar/opt/fastx/latest/bin/fastx-protocol

   

4. Click OK.

5. Select the remote desktop collection (double-click or press Enter).
6. Authenticate with your username and token response.

7. Click the + button in the upper-left corner of the next window.
   

8. Select the KDE icon that displays startplasma-x11 in the required command field.

   1. Double-click the icon to start the KDE desktop. You can then right-click on the desktop to get a Kconsole menu, OR
   2. Double-click the xterm icon instead to get a single xterm.