Aalto provides some Linux computers which can be used for training.
Paniikki is one of Aalto's computer labs for students. The computers can be accessed remotely using SSH. macOS comes with an SSH client preinstalled, but on Linux and Windows you may have to install it yourself:
The hostnames of the Paniikki computers can be found on Aalto's Linux computer names in IT classrooms page.
In the following examples we will use the username myusername and the computer known as whitespace. In practice you should substitute your own username and look through the available computers to find one that's not being used (more on that later).
Connect using the ssh command and you will be prompted for your password. Type it in (you won't see any output while typing) and press enter to log in.
If you're on the Aalto network, you can connect directly by running:
ssh myusername@whitespace
If you're not on the Aalto network, you need to "jump" via one of the shell servers (kosh or lyta) that are accessible from outside. For example:
ssh -J [email protected] myusername@whitespace
You'll have to enter your password twice in this case.
After logging in, you will be presented with a Linux shell prompt. If you're new to the Linux command line, you may want to check out a tutorial such as this one.
When you want to log out, run:
logout
You probably don't want to run heavy calculations on a computer that others are already using. To check for activity, you can use:
htop
This shows a live monitoring view of things like CPU load and used memory. If most of the CPU cores are active or most of the memory is used, you should try another system (check the list linked above). Press q to exit htop.
You can also check live GPU utilization by running:
watch -n 1 nvidia-smi
This runs the command nvidia-smi every second. Press ctrl-C to exit watch. (In fact ctrl-C can be used to exit most interactive processes.)
To use Conda on Paniikki computers, you first need to load the anaconda3 module using Lmod, like so:
module load anaconda3
This applies for the remainder of your login session and you will need to run it again if you log out and later log back in.
See the Lmod manual for more commands.
After this, you should be ready to set up Conda environments as described in our other instructions for Linux, and to run training scripts etc. on the remote system.
Note: If you get an error while creating a conda environment, you might have to run
conda init bash
then log out and log back in and run
source ~/.bashrc
And then rerun the conda environment creation command.
Training the neural networks covered in the course can take anywhere from a few hours to ~5 days. Normally, whatever processes you're running will be terminated if you log out, get disconnected, etc. To avoid this, you can use the screen "terminal multiplexer" utility to leave a process running in the background.
First, start a new screen by running:
screen
This will show a startup message. Press space or enter to dismiss it and you'll get a new shell prompt. Anything you run here will persist even if you're not connected to the remote system.
At this point, you would typically load your Conda environment and start whatever script you want to run.
Note: If using screen, you should load the anaconda3 module (see above) after starting screen!
To detach from the screen, i.e. exit it while leaving it running in the background, press ctrl-A followed by ctrl-D.
You can list your running screens by running:
screen -ls
This will output something like:
There is a screen on:
1465521.pts-14.unlambda (2021-04-06 16:47:34) (Detached)
1 Socket in /run/screen/S-kastemm1.
To reattach the screen again, run:
screen -r
If running multiple screens, you have to tell screen -r which one to reattach by giving it the screen's ID, e.g.:
screen -r 1465521
You might have made a typo when specifying the ID, or the screen is already attached in another session. A screen may be left attached if your network suddenly disconnected while working in the screen. You can force it to detach first by using the -D flag, e.g.:
screen -Dr 1465521
In summary, to run e.g. a training script on a Paniikki computer:
- Log in with
ssh - Start a
screen - Load the Anaconda module —
module load anaconda3 - Activate your Conda environment —
conda activate ... - Run the script —
python ...
When doing long runs, it can be easy to forget which computer you chose, so we recommend making a note of this.
You can transfer files to and from your Aalto home directory using SMB, but you need to be on the Aalto network to do this. When working remotely, you can use the Aalto VPN to join the network.
Once on the Aalto network, you can connect to the SMB server.
In Finder, open the Go menu and select "Connect to Server...". Enter:
smb://home.org.aalto.fi
Click Connect and log in with your Aalto credentials.
If you prefer not to set up the Aalto VPN, there is another method on Mac: you can create an SSH tunnel via an outward–facing Aalto shell server like kosh or lyta.
Open a separate terminal window and run: (replacing myusername with your Aalto username)
ssh [email protected] -L 127.0.0.1:10139:home.org.aalto.fi:139
In Finder, you can then open the Go menu, select "Connect to Server..." and enter:
smb://localhost:10139
Click Connect and log in with your Aalto credentials.
The tunnel will be closed when you log out of the ssh connection.
Open an Explorer window, and enter in the address bar:
\\home.org.aalto.fi
Press Enter and log in with your Aalto credentials.