INMA Documentation
Toggle Dark/Light/Auto mode Toggle Dark/Light/Auto mode Toggle Dark/Light/Auto mode
  1. Welcome
  2. /
  3. Application server

Application server

An application server has been funded by Philippe Lefèvre. If necessary, the researchers of his group have priority, but normally it is accessible to all the researchers of the INMA pole.
No backup is made on this machine. It is up to you to keep a copy of your programs and the results obtained on your personal machines.

Here is some useful information.

  • It is a Dell Precision 5820 (128 GB RAM, 18-core processor and a Nvidia Quadro RTX6000 GPU graphics card with 24 GB for calculations using the GPU). The OS is Ubuntu 20.04 LTS.

  • The server name is anatole.inma.ucl.ac.be and its IP is 130.104.236.1. The access is done by SSH only from the UCLouvain network. Concretely, if you work from home, you must use a VPN connection to the university intranet.

  • For researchers, the access authorization to the VPN is done via the service catalog in the web interface of the service desk : https://servicedesk.uclouvain.be/ServiceDesk.WebAccess/ss/. For graduate students, this request must be made by the supervisor.

  • The installation of the VPN client is described on https://intranet.uclouvain.be/fr/myucl/services-informatiques/vpn-openvpn.html. In case of problems with the VPN connection, please contact the service desk. They are the ones who manage this service.

  • Depending on the OS of your personal machine, you will have to install an SSH client and possibly an X server to display locally the application windows running on the server. For Windows, I recommend MobaXterm which does both. There is a free version that you can download, but UCLouvain has a site license that gives access to a version with more functionality. If you wish, contact me by mail (etienne.huens@uclouvain.be) so that I can give you access to it.

  • There is no system to manage the sharing of resources between users. We ask you to be reasonable. The server can run up to 36 threads simultaneously, depending on the number of users, so share equally. If this is not enough, it is possible to access the university’s HPC service (https://uclouvain.be/en/research/cism). If you use more than one thread when running a program, implement it so that it is a variable. You will be able to adapt them easily.

  • Matlab R2020a, CUDA, Tensorflow, TensorRT, R and opencv are installed. For Python, install Anaconda in your home directory. If you need other software, contact me (etienne.huens@uclouvain.be).

Implement your programs so that the intermediate results are saved regularly and that your programs can be restarted in case of unexpected stop. It also allows me to do a reboot to apply updates without impacting your work.
Depending on the number of users, I may contact you to ask you to reduce the number of threads used.

Let’s say you need to run some code performing very long computations on the anatole server, within a remote session. If you disconnect your terminal from the remote session before those computations are completed you will see that all your processes currently running on the remote server will get killed, which is bad :-(

Here is a solution to avoid that, using the utility screen

  1. Before you start your long computations type in the remote terminal screen
  2. Press the spacebar, a new window opens. You can now launch the code that performs your computations.
  3. Once your code is running you may now “detach” from the screen session by pressing ctrl-a d which will bring you back to where you were right before you started the screen utility
  4. Don’t worry, your code is still running inside of the now detached screen session. You can also quit the remote session now.
  5. When you want to reconnect to the terminal with your code running, open a new remote terminal on anatole and type screen -R. This will reconnect you (reattach) to the still-running terminal. You can then detach again if you wish.
  6. When you are done with your computations you can exit the screen session using exit. Don’t forget to do this: without it, the screen process continues to run unnecessarily on the server.

Optionally, you can create several screen sessions (press ctrl-a c inside of screen). You can then list all currently screens with screen -ls and even select which session you reattach to with screen -r [session_name].

If your code runs inside of a graphical interface (X11) disconnecting from your session may cause problems (I haven’t found a way to avoid them). In case you want to run MATLAB code the solution consists in running MATLAB without a graphical display, using matlab -nodisplay