Launching CONN
CONN is a graphical application. It needs a display, which means you cannot
simply ssh in and start it — and you should not run it on a login node in
any case.
There are two ways to work, and which you pick depends on whether you need to see the interface.
| Interactive session | Batch job | |
|---|---|---|
| You get | The CONN GUI on screen | No interface at all |
| Use it for | Building a project, inspecting results | Reproducible, unattended runs |
| Covered in | This page | Submitting Jobs from MATLAB |
Interactive Session
Use SCC OnDemand, which gives you a graphical session on a compute node through your web browser — no SSH client and no X11 forwarding.
- Go to scc-ondemand.bu.edu and log in with your BU credentials.
- Under Interactive Apps, choose the Desktop app.
- Fill in the session request: your project, the number of cores, and how many hours you need.
- Launch the session, then connect to it when it is ready.
Once the desktop appears, open a terminal inside it and load the three modules in this order:
[scc4]$ module load matlab/2024b
[scc4]$ module load spm/25.01.02
[scc4]$ module load conn/25b
Load them in the terminal, not in the session request. The Desktop app starts a plain graphical session with no software loaded, which is why the module commands come after you connect rather than before. Out of order, the CONN module refuses to load and tells you what it needs — see Getting Started.
Then start MATLAB from the same terminal:
[scc4]$ matlab &
and start CONN from the MATLAB command window:
>> conn
How many cores should I ask for?
Batch Job
If you do not need to see the interface, run CONN from a conn_batch script
submitted to the queue. MATLAB runs with no display, the job runs unattended,
and the script is a reproducible record of what you did.
That is covered in Submitting Jobs from MATLAB.
Do Not Run CONN on a Login Node
Login nodes are shared by everyone on the SCC and are not for computation. Starting MATLAB and running a pipeline there will slow the node down for every other user, and RCS may terminate the process.
Use an OnDemand session or submit a job.