Remote desktops and visualization tools
Remote desktops are a tool designed for accessing FinisTerrae III in a visualization mode that is more user-friendly for certain users. To create a remote desktop, navigate to the User Portal > Tools > Remote visualization FT-III. There is no need to use the VPN to connect to the remote desktops, making this tool very useful in the event of potential VPN connection issues.
Once you have created a desktop, it will remain active for a maximum of 36 hours. If you do not extend the duration of your session, the desktop will automatically close, and any unsaved data will be lost.
To monitor the remaining time for your desktop, check the top bar. By clicking on it, you can extend your session for an additional 36 hours. Please note that changes may take up to 2 minutes to become effective and will not be immediate.
If you slide your mouse over the circular graphic icon, you can view your Home quota limits, including those for file and storage space.
The screen icon is intended to adjust the resolution of the remote desktop to match that of the user’s local screen.
Terminal icon will open a new terminal window at user’s Home directory of the FinisTerrae III.
Firefox icon will launch a new window and open the website.
Clipboard: is the left panel used to facilitate communication between your local PC and the remote desktop. It includes a keyboard, a clipboard that allows copying and pasting between your local PC and the remote desktop, a full-screen mode, settings and a disconnection icon.
If you experience issues while opening a Remote Desktop from both the User Portal and SSH sites, it may be due to exceeding your $HOME quota or approaching the limit. In such cases, you will not be able to create a Remote Desktop. To solve this issue, free up space in your $HOME directory by deleting any unnecessary data or moving it to your $STORE or $LUSTRE directories. For further information about storage and quota limits, please refer to: Permanent Storage.
FinisTerrae III also provides a desktops CLI that you can use to manage your remote desktops. The CLI is preconfigured with bash autocompletion and supports the following commands:
desktops: allows you to manage your remote desktops from the command line.
desktops create: creates a new desktop and returns the URL.
desktops delete: erases the desktop. This command can be used as a backup option to delete the desktop if the option available on the User Portal is not working for any reason.
desktops extend: extends the time for the remote desktop by 36 hours.
desktops --help: provides additional information on how to use the command.
Jupyter Notebook environment is presented in the form of a web document that combines text, code, and visualizations. This makes it an ideal tool for data exploration and visualization, as well as for creating interactive reports and presentations. It is installed as a base module in our system, and there are several ways to use it.
Remote desktops: As previously explained, you can launch a remote desktop and use Jupyter Notebook there. It is important to note that remote desktops run on a GPU T4 node, so your jobs will be executed on it. To run Jupyter, use the command
jupyter-notebook. The hostname will be automatically completed and you can enter jupyter using the link created by the system (if not launched automatically). If the connection is denied by permissions, run jupyter using
jupyter-notebook --ip $(hostname -i).
Terminal access: you can run Jupyter by command line without the need for remote desktops. When you connect to FinisTerrae III, you are automatically redirected to a login node, but these nodes are not for computing. Based on the type of resources you need (GPU: a100 or T4, memory, visualization…), there are different ways to run Jupyter notebook:
Using a visualization node: to get access to a GPU T4 node, run command
compute --gpu. You will be granted the resources at a time and, to run Jupyter use
jupyter-notebook --ip $(hostname -i). The hostname will be automatically completed and you can enter Jupyter using the link created by the system (if not launched automatically).
- Using a GPU a100 node: if you specifically need a GPU A100 node, you can make a request in two ways:
salloccommand: salloc –gres=gpu:a100:1 -t 00:10:00 –mem=2GB -c 32. Number of cores may differ between the requested nummber of GPUs, for further information visit this link. Once the resources are provided, there will be a running job in the node, and to run Jupyter, you must first access the node by SSH:
ssh <node_name>. Once connected, run
jupyter-notebook --ip $(hostname -i)as explained above.
srun: This is similar to “salloc,” but in this case, you won’t need to access the node, making it an easier way. Again, when the resources are acquired, run
jupyter-notebook --ip $(hostname -i).
Using a ilk or smp node: this is similar to running it in a GPU a100 node but, in this case, you don’t need to add the –gres=gpu:a100:1 option. Without this option, the system will grant you access to an ilk or smp, depending on the RAM memory requested.
Unless you are accessing from a remote desktop, all other connections will require the VPN to be active at all times, even if you are connected from an authorized center. This is because the connection is made to a private address and in order to access it, the connection must come through the VPN tunnel.