IT documentation

0. Content

1. GPU/CPU Computation Resources

1. GPU/CPU Computation Resources: icclusters (iccluster.epfl.ch)

1.0 Logins

current available clusters: 047(2 GPUs)

Logon to the server: ssh {your gasper ID}@iccluster{047}.epfl.ch

1.0.1 Create a shortcut for SSH:

add the following lines into your ssh config file (usually it’s ~/.ssh/config or /etc/ssh/ssh_config)

Host 089

    Hostname iccluster089.iccluster.epfl.ch

    User {your gasper ID}

Then you can simply use ssh 089 instead of ssh {your gasper ID}@iccluster089.iccluster.epfl.ch

1.1 File system and File Management

Basic philosophy: save configure files in homedir, put temporal code/data for your experiments in /scratch, and backup the code (and the data you collected/ not available online) in /ivraldata1

1.1.1 Homedir

location: /home/{your gasper ID}, has a small volume of 25GB, is shared among clusters, thus you can put your configure files here (for example .bashrc, .vimrc, .emacs, conda etc.). Only you have the write permission in your homedir.

Note: chmod/chown and symlink/hardlink is not available in homedir

1.1.2 /scratch

location: /scratch, you need to create your own folder and work inside the folder. You will be the owner of the folder you created and other people won’t have the write permission unless you give them (for example by using chmod). /scratch has a very big volume of 25TB, is locally mounted on each cluster, the data in it is not backedup. Thus you should only put temporal code/data in this folder.

If you are doing the experiments on the same data/code on different clusters, you should still put your data in /scratch although it’s not shared among clusters, you can use command scp for copying data from one cluster to the other.

Supervisors have the responsibility to clean the students’ folder when the projects finish.

1.1.3 /ivrldata1

location:/ivraldata1, it is for collaborative share and for storing all codes and data generated by the lab. Now it has a volume of 2TB, and can be upgraded up to 25TB (165CHF per TB per year).

Folder hierarchy:

  • /ivrldata1/ivrl, for storing common data for IVRL, IVRL staff has read/write permission on it
  • /ivrldata1/home, you need to create your own folder and put your research data/code (after cleaning) here. Only the creator of the folder has the write permission to the folder.
  • /ivrldata1/data, you need to create your own folder and put your created/collect datasets (after cleaning) here. Only the creator of the folder has the write permission to the folder.
  • /ivrldata1/students, it is the workspace for students, creator of folder has the write permission. Francoise can give login permission to the students https://groups.epfl.ch/viewgroup?groupid=S14694

Supervisors have the responsibility to clean the students’ folder when the projects finish.

1.2 Reserve and setup a New Node

1.2.1 Reserve a New Node

  1. go to https://install.iccluster.epfl.ch/Portal/, login with your gasper account
  2. go to Reservations->Make a reservation
  3. select your days (normally you can modify it later) and server type (normally we reserve 1 ICCT3, which has 2 Titan X)
  4. setup your node once it appears on My Servers->List

1.2.2 Setup a New Node

  1. go to https://install.iccluster.epfl.ch/Portal/, login with your gasper account
  2. go to My Servers->Setup
  3. add the new reserved node into the setup list
  4. choose a boot option, normally select Ubuntu xenial (16.04)
  5. in Customization, select at least Add IC-IT SSH Keys to root and IVRL Customization
  6. in Run setup, click I confirm
  7. wait for around 2-3 hours after the boot is finished

1.3 Softwares

https://c4science.ch/source/ivrl_conf/