4. Operation of Open OnDemand¶
This section describes how to use Open OnDemand.
4.1. Screen layout and menu bar¶
This section explains how to use the menu bar at the top of the screen.
menu bar
4.1.1. File operations¶
Select Home Directory from Files on the menu bar to display TSUBAME's home directory.
Standard file operations can be performed from the buttons on the screen.
| Button | Operation | Description |
|---|---|---|
| Open in Terminal | Terminal | Open TSUBAME terminal with a new tab |
| Refresh | Refresh | Refresh directory |
| New File | File | Create new file |
| New Directory | Directory | Create new directory |
| Upload | Upload | Upload a file |
| Download | Download | Download selected file(s) |
| Copy/Move | Copy/Move | Copy/Move selected file(s) |
| Delete | Delete | Delete selected files(s) |
| Change directry | Directory | Change directory |
| ↑ | Directory | Move up to the parent directory |
| Copy path | Path | Copy current directory's path |
4.1.2. Login to TSUBAME¶
Selecting "Clusters > TSUBAME cluster Shell Access", you can login to TSUBAME with. Multiple terminals can be opened.
To log out of TSUBAME and exit the terminal, type exit and close the browser tab.
Tips
The terminal will automatically disconnect after a certain period of inactivity. Please click here for a workaround. Session suddenly disconnected while working on TSUBAME4.0
4.1.3. Interactive Apps¶
For Interactive Apps, lease refer to 3.2. Interactive Apps.
4.1.4. Sessions¶
My Interactive Sessions can display own sessions.
Press Delete for each session to exit the session.
Tips
When the browser is narrower, the My Interactive Sessions string in the menu bar disappears, leaving only the icon display.
4.1.5. Help and Restart¶
Press Help for manual display and refresh.
If you have problems with Open OnDemand, try Restart Web Server first.
From "Access Management Page", you can login to Access Management Server.
4.1.6. Logout¶
To log out of Open OnDemand, press Log Out on the menu bar.
Warnning
If you close your browser without pressing Log Out, your Open OnDemand session will remain open for a certain amount of time. If you close your browser without pressing Log Out, your Open OnDemand session will remain for a certain period of time, which may cause problems when you try to log in to Open OnDemand again.
4.2. Interactive Apps¶
The following three Interactive Apps are available for Open OnDemand in TSUBAME4.0.
- TSUBAME Desktop (xfce)
- Jupyter
- code server
Interactive Apps are further divided into two types according to execution type.
- Use normal compute nodes
- Use a dedicated queue for interactive jobs (* interactive is included in the name)
Differences by execution type are as follows
| Execution type | Use normal compute nodes | Use Interactive job queue |
|---|---|---|
| Mode of Use | onerous | Free on-campus use , onerous |
| Resource type | selectable | Interactive job queue only (Assigned resources number of physical CPU cores 24 cores, 96GB memory, 1MIG, but up to 12 people share the same resources.) |
| Number of nodes | selectable | 1 node only |
| Premium Options | available | not available |
| Reserving | available | not available |
Below are the instructions for using Interactive Apps.
Info
The screenshots shown here are for "TSUBAME Desktop (xfce)".
There is no difference in the settings depending on the application, so please read them accordingly.
4.2.1. Selecting the application/execution type¶
From Pinned Apps at the bottom center of the screen, click the icon for the application you wish to use.
Go to the environment setting screen.
4.2.2. Environment setting¶
Set up the environment items and click the Launch button. The request is submitted as a job.
| Catption | Item | Description |
|---|---|---|
| Select resouce type | Resource type | Select required resource type |
| Number of Request resources | number | Required node number |
| TSUBAME group | group | Select TSUBAME group |
| Maximum run time(hh:mm:ss) | duration | Set the maximum value for the scheduled usage time (can be terminated midway) |
| Priority Option | priority | Priority |
| Reservation Number (AR ID) | ARID | Specify the reservation number (AR ID) when using a reservation |
Info
When using Interactive job queue, "TSUBAME group" and "Maximum run time" can only be set.
Info
Interactive job queue is available free of charge only for intramural users (tgz-edu) and access card users.
To use Interactive job queue free of charge, please specify “Free” in the “TSUBAME group” field.
Please note that if you specify a group, you will be charged for the target group.
Tips
The only resource types for which reservations can be used are node_f, node_h, node_q, and node_o.
4.2.3. Launching Applications¶
The request is submitted as a job, and when the specified resource is allocated, the session is set to Running.
If it is Queued, it is waiting to be executed, so wait a while before checking the status.
Once the session is Running, click “Launch app name” to launch the application.
Info
Compression” and ‘Image Quality’ are only displayed in TSUBAME Desktop (xfce).
Info
Pressing View Only (Share-able Link) opens a copy of the X desktop in a new tab.
You will not be able to control the screen, but you will see a mirrored display of the previous desktop.
By sending the URL of this tab via e-mail or other means, you can simply share or monitor the screen.
Warning
OpenOnDemand will reserve computing resources even when the application is not in use, as long as the session is running, and will be eligible for TSUBAME Point payments.
From the viewpoint of effective use of resources, please delete the session manually when you finish using the application.
Below are the initial screens of each application (there is no difference depending on the execution type).
Please refer to the official website for details on how to use these applications.
TSUBAME Desktop (xfce)
official website : https://www.xfce.org/?lang=en
Jupyter
official website : https://jupyter.org/
code server
official website : https://coder.com/docs/code-server
4.2.3.1. Registering a virtual environment on Jupyter (ipykernel)¶
On the kernel (ipykernel) available on Jupyter, it is not possible to execute module load.
To solve this problem, we introduce a method to create and register a virtual environment with necessary modules loaded.
- Creating a virtual environment with loaded modules
- Register the virtual environment to Jupyter (ipykernel)
- Delete the registered virtual environment from Jupyter
Info
Verified with Jupyterlab 4.1.4 / CUDA 12.8.0 / python 3.9.18 combination.
4.2.3.1.1. Creating a virtual environment with loaded modules¶
The following are the steps for creating a virtual environment with the module loaded.
Info
If you use OpenOnDemand's Jupyter Terminal, modules such as jupyterlab are loaded at login. Be sure to execute the module purge command before creating the virtual environment.
In the example in this section, the name of the virtual environment (virtual_environment_name) is temporarily set to env01. It can be changed to any name.
1.Create and activate an empty virtual environment
[login/rNnN]$ module purge
[login/rNnN]$ python -m venv virtual_environment_name # Replace virtual_environment_name with the name of your virtual environment.
[login/rNnN]$ source virtual_environment_name/bin/activate # Replace virtual_environment_name with the name of your virtual environment.
< Example >
[tsubame_user@r8n10 ~]$ cd ~
[tsubame_user@r8n10 ~]$ module purge
[tsubame_user@r8n10 ~]$ python -m venv env01
[tsubame_user@r8n10 ~]$ source env01/bin/activate
(env01) [tsubame_user@r8n10 ~]$
Info
After activating the virtual environment, the name of the specified virtual environment will be displayed at the prompt.
You can now add packages, load modules, etc. to the virtual environment using pip / module load, etc.
To exit from the virtual environment, execute the deactivate command.
< Example >
(env01) [tsubame_user@r8n10 ~]$ deactivate
[tsubame_user@r8n10 ~]$
2.Load the required modules.
Load modules in the virtual environment you have created.
As an example, load CUDA modules to use GPUs, etc.
[virtual_environment]$ module load cuda/12.8.0
< Example >
(env01) [tsubame_user@r8n10 ~]$ module load cuda/12.8.0
(env01) [tsubame_user@r8n10 ~]$ module list
Currently Loaded Modulefiles:
1) cuda/12.8.0
Key:
default-version
4.2.3.1.2. Register the virtual environment to Jupyter (ipykernel)¶
Info
In the example in this section, the name of the virtual environment (virtual_environment_name) is temporarily set to env01. It can be changed to any name.
Register the virtual environment created by Creating a virtual environment with loaded modules using a dedicated tool (t4jpyregister).
[virtual_environment]$ /apps/t4/rhel9/utils/bin/t4jpyregister -n virtual_environment_name # Replace virtual_environment_name with the name of your virtual environment.
< Example >
(env01) [tsubame_user@r8n10 ~]$ /apps/t4/rhel9/utils/bin/t4jpyregister -n env01
INSTALL TO ${HOME}/.local/share/jupyter/kernels/env01
On the Launcher screen of Jupyter, you can confirm that env01 has been added (the red frame part. It takes some time to reflect).
To check on the command line, execute the following command after loading the jupyterlab module.
Info
Please execute it with deactivated from the virtual environment.
[login/rNnN]$ module load jupyterlab/4.1.4
[login/rNnN]$ jupyter kernelspec list
< Example >
[tsubame_user@r8n10 ~]$ module load jupyterlab/4.1.4
[tsubame_user@r8n10 ~]$ jupyter kernelspec list
Available kernels:
env01 ${HOME}/.local/share/jupyter/kernels/env01
python3 /apps/t4/rhel9/free/jupyterlab/4.1.4/gcc11.4.1/python3.9.18_b/share/jupyter/kernels/python3
4.2.3.1.3. Delete the registered virtual environment from Jupyter.¶
To delete the registered virtual environment from Jupyter, execute the following command after loading the jupyterlab module.
Press “y” when you are prompted to confirm that you want to delete the file.
Info
The virtual environment file itself will not be deleted.
Please execute it with deactivated from the virtual environment.
[login/rNnN]$ module load jupyterlab/4.1.4
[login/rNnN]$ jupyter kernelspec uninstall virtual_environment_name # Replace virtual_environment_name with the name of your virtual environment.
< Example >
[tsubame_user@r8n10 ~]$ jupyter kernelspec uninstall env01
Kernel specs to remove:
env01 ${HOME}/.local/share/jupyter/kernels/env01
Remove 1 kernel specs [y/N]: y
Removed ${HOME}/.local/share/jupyter/kernels/env01
env01 has been removed from the Jupyter Launcher screen.
4.2.4. Deleting a session¶
To exit the X desktop, close the browser tab and Delete the session from the session list. Once closed, the session will be marked as Completed.
The session will also be marked as Completed when the specified usage time has elapsed.
Sessions that have been marked as Completed will be automatically deleted from the list after a certain period of time.
