cmslab-workstation

The private server of cmslab.

General Specifications

see detailed specifications here

Workflows installed

JupyterHub

JupyterHub is installed as a docker image. It is modified it to match the lab's cases. The name of the image is fixedjpyhub:ver2 . Use this command to list all the docker images installed:

docker images

To start the JupyterHub service run:

docker run -p 8000:8000 -d --name CMSLABjupyterhub fixedjpyhub:ver2 jupyterhub

Access the JupyterHub server by going to http://10.4.48.61:8000 (opens in a new tab) on your browser.(or whatever the current IP address is) \ To add new users first get into the container bash shell:

docker exec -it CMSLABjupyterhub bash

The container is has debian as the linux layer. So, add a new user by :

adduser <your username>

Machine Learning with GPU

A singularity container with tensorflow and jupyter is installed which utilises the Nvidia GPU installed in the server.

To start a jupyter notebook session with tensorflow and GPU :

1. Forward the jupyter notebook port 8888 via ssh.

ssh -L 8888:localhost:8888 <username>@10.4.48.61 

2. Start the singularity container with -nv flag to allow the usage of Nvidia GPU :

singularity run --nv /commonfs/singularity_images/ML/tensorflow_latest-gpu-jupyter.sif

3. Ctrl+Click on the http://127.0.0.1:8888 … link which shows up on the terminal.

Jupyter-Lab

Natively, the jupyter-lab installed on bare metal of the server has no additional packages installed. Although you could use pyROOT. Additional packages could be installed for a user by adding the -user flag during pip installations.

To start a jupyter-lab server, first login to the server with an appropriate port forward. For this example lets forward the port 8934 :

ssh -L 8934:localhost:8934 <username>@10.4.48.61

Once that is done, lets start a jupyter-lab server at port 8934 :

jupyter-lab --no-browser --port 8943 --ip=localhost

Upon running this, you should see two links on the terminal. Ctrl+Click on the link that starts with https://127 ...

This will open your browser and lead you to the jupyter-lab console. Here you could use pyROOT by

import ROOT

Expected output is a welcome message from JupyROOT.

Other information

Life hacks

1. LAN failure
   In the case when university LAN internet certification expires you can still login through the network, follow these steps:

   1. Login as usual with ssh but with X-forwarding enabled:

   ssh -X <username>@10.4.48.61
   2. Open a lightweight browser and login to the university network. A lightweight browser called seamonkey is already installed and it is recommended to use that: To start the browser:

   seamonkey
   3. The university login url is http://192.168.56.2:8090/httpclient.html (opens in a new tab) which is also bookmarked in seamonkey. Once logged in, you could close the browser window.

2. In the case when the university LAN has no power We have setup an internal switch network with the 'Bhawna 2' wifi router. In case of a university LAN failure,

   1. Connect to Wifi on your device and connect to 'Bhawna 2' (Password: CMSLAB2@).
   2. Go on to the settings of the network 'Bhawna 2'
   3. Change the ipv4 and ipv6 settings to 'Link Local' and apply the changes.(One-time process)
   4. Disconnect and reconnect to 'Bhawna 2'
   5. Open a terminal and ssh connect to the server with your username and ip address: 169.254.137.50 (this is the hardware ip4).
      For example :

   ssh prayag@169.254.137.50

   Known issues

3. IP address assignment
   — prayag 2024/02/05 06:48
   It seems like the IP address assignment by university network has a problem. I have setup the NetworkManager configuration to assign 10.4.48.61 to the device id eth0 . This assigns the ip 10.4.48.61 to the server on startup. But this work almost 50% of the times only. Sometimes, the IP address 10.4.49.35 is assigned on startup. For now, this could be fixed by repeatedly rebooting the server till 10.4.48.61 is assigned.

4. Logging in to university wifi
   — prayag 2024/02/11 11:24
   Once you connect to either University LAN or wifi you have to login from a web-browser. The problem with a server configuration is that you can't always have an access to the GUI, to open a browser and login from a browser. A possible solution to this is re-engineering the AJAX request from the login portal and run the request from a terminal. I have tried this and have been unsuccessful so far. A workaround is to open a VNC session and login from there. You can also “X forward” a firefox browser window through ssh but that is incredibly slow. A lightweight browser like seamonkey will work a lot better. Go to seamonkey and check out the bookmark where the university wifi login page is bookmarked.