TACC Analysis Portal

The TACC Analysis Portal (TAP) is a web-based portal that provides access to interactive sessions on TACC HPC systems. TAP-supported applications include Desktop Cloud Visualization (DCV) remote desktop, Jupyter Notebooks, and RStudio. Any user with an allocation on one of TACC’s HPC systems can log in to TAP and use the services. Over the course of this workshop, we will be using TAP to launch Jupyter Notebooks and DCV remote desktop sessions as a convenient way to interatively write/run Python code and build/train machine learning models. By the end of this section, you should be able to:

Log in to the TACC Analysis Portal
Submit a Jupyter Notebook job to a TACC system
Connect to a Jupyter Notebook session

Accessing the Portal

Log in to TAP using the same username and password that you use to access the TACC User Portal. Once you’ve logged in you’ll be directed to the Home Screen where you can begin scheduling jobs.

Job Management

Submitting a Job

Submitting a job on TAP requires the following inputs: ( 1 )

System: where the job will run. The system selector drop-down will contain the TAP-supported TACC systems where you have an allocation. The system must be selected first, as the values of the other selectors are determined by the selected system.
Application: which application the job will run. The application selector will contain the applications available on the selected system (DCV, VNC, Jupyter, or RStudio)
Project: which project allocation to bill for the job run. The project selector will contain the projects associated with your account on the selected system.
Queue: which system queue will receive the job. The queue selector will contain the TAP-supported queues on the selected system.
Nodes: the number of nodes the job will occupy. We recommend leaving this setting at 1 unless you know you need more nodes. This is equivalent to the -N option in SLURM.
Tasks: the number of MPI tasks the job will use. We recommend leaving this setting at 1 unless you know you need more tasks. This is equivalent to the -n option in SLURM.

A TAP job also accepts these additional optional inputs: ( 2 )

Time Limit: how long the job will run. If left blank, the job will use the TAP default runtime of 2 hours.
Reservation: the reservation in which to run the job. If you have a reservation on the selected system and want the job to run within that reservation, specify the name here.
VNC Desktop Resolution: desktop resolution for a VNC job. If this is left blank, a VNC job will use the default resolution of 1024x768.

After specifying the job inputs, select the Submit ( 8 ) button, and your job will be submitted to the remote system. After submitting the job, you will be automatically redirected to the job status page. You can get back to this page from the Status ( 3 ) button. If the job is already running on the system, click the Connect ( 5 ) button from the Home Screen or Job status to connect to your application.

Click the “Check Status” button to update the page with the latest job status. The diagnostic information will include an estimated start time for the job if Slurm is able to provide one. Jobs submitted to development queues typically start running more quickly than jobs submitted to other queues.

Ending a Submitted Job

When you are finished with your job, you can end your job using the End ( 4 ) button on the TAP Home Screen page or on the Job Status page. Note that closing the browser window will not end the job. Also note that if you end the job from within the application (for example, pressing "Enter" in the red xterm in a DCV or VNC job), TAP will still show the job as running until you check status for the job, click "End Job" within TAP, or the requested end time of the job is reached.

Resubmitting a Past Job

You can resubmit a past job using the Resubmit ( 7 ) button from the Home Screen page. The job will be submitted with the same inputs used for the past job, including any optional inputs. Select Details ( 6 ) to see the inputs that were specified for the past job.

Utilities

TAP provides certain useful diagnostic and logistic utilities on the Utilities page. Access the Utilities page by selecting the Utilities ( 9 ) button on the Home Screen page.

Configuring Jupyter Notebook

The Utilities section provides access to several common actions related to Jupyter Notebooks.

“Use Python3” sets the TACC Python3 module as part of your default modules so that TAP will use Python3 for Jupyter Notebooks. If you want to use a non-default Python installation, such as Conda, you will need to install it yourself via the system command line. TAP will use the first jupyter-notebook command in your $PATH, so make sure that the command which jupyter-notebook returns the Jupyter Notebook you want to use. Conda install typically configures your environment so that Conda is first on your $PATH.
“Link $WORK from $HOME” and “Link $SCRATCH from $HOME” create symbolic links in your $HOME directory so that you can access $WORK and $SCRATCH from within a Jupyter Notebook. TAP launches Jupyter Notebooks from within your $HOME directory, so these other file systems are not reachable without such a linking mechanism. The links will show up as “WORK” and “SCRATCH” in the Jupyter file browser. You only need to create these links once and they will remain available for all future jobs.

Warning

Vista by default does not have the Python3 module loaded, nor is it in the default module path. This means that if you try to run a Jupyter Notebook on Vista, it will return an error. To fix this, you must manually log in to Vista on the command line and perform the following steps:

[vista]$ module load gcc
[vista]$ module load python3
[vista]$ module save

This is a one-time setup step. After this, you can use TAP to launch Jupyter Notebooks on Vista.

Obtaining TACC Account Status

The Status section provides system information and diagnostics. “Check TACC Info” will show account balances and filesystem usage for the selected system. “Run Sanity Tool” performs a series of sanity checks to catch common account issues that can impact TAP jobs (for example, being over filesystem quota on your $HOME directory).

Troubleshooting TAP

If you encounter an error while using TAP, or if your job status remains in “PENDING” for more than a few moments, alert the workshop instructors.

Note

For this workshop, we will be using Frontera and Vista systems, and we will be launching both DCV sessions and Jupyter Notebooks. Each day has a unique reservation name. Make sure to choose the right system, application, and reservation name each day. If you are not sure, ask one of the instructors for help.