How to get server access?

Generate a public key using ssh-keygen -b 4096 and send the file ~/.ssh/id_rsa.pub to Florian Teich (with ~ being your home directory). After your account has been created you can access it by the command ssh <username>@frankfurt.physik3.uni-goettingen.de or ssh <username>@134.76.92.27.

For comfortable access you might want to create a ~/.ssh/config which looks like this:

Host frankfurt
    HostName frankfurt.physik3.uni-goettingen.de
    Port 22
    User <username>

Please take care not to waste any resources and interfere with other users.

Rules

• Some general infos
  • Linux Ubuntu 18.04 LTS, intel 40 cores @ 2.6 GHz, 344 GB memory
  • IP address: 134.76.92.27
  • Login: only via SSH terminal, only via Private Key
    • Do not loose them
    • If lost, send me mail and i will add new key and remove the old one
    • Only via GWDG VPN
• Limited Ressources
  • Please do not exceed max RAM (in Person hard reboot required otherwise, which is hard to do especially during lockdown)
  • Use CPUs wisely
    • 40 Cores, but usually every user needs some cores, so please do not use all of them
  • Use GPUs wisely
    • Use Environment Variables and monitor Memory and Computation use of GPUs via nvtop or nvidia-smi
    • Check GPU Usage via nvtop
      • Optimize your scripts, especially your dataloaders to not congest the GPUs (use num_workers etc.)
        • indicator for bad management: low gpu computation but high gpu memory usage
    • Use Collaborative Pad for letting everyone know that you need ressources (see link below)
  • Use Hard Drives wisely
    • Use only bare minimum in home directory
    • Save datasets you do not currently use on HDDs
    • Save datasets you currently use preferrably on SSDs
  • Shared OS
    • If you need certain applications that are not yet installed system wide but may benefit multiple users: talk to me
    • If you do not have certain applications/libraries:
      • Compile and/or install them inside your own userspace
      • Use environment managers (i.e. pip for python)
      • Use binary manager (i.e. anaconda)
      • Use virtualization (i.e. docker-io)

Important for GPU Users

• If possible, use only one GPU at a time:

export CUDA_VISIBLE_DEVICES="0"

• Link to the GPU-Usage Pad: https://pad.gwdg.de/pvle6JQjQzep1Urx4VZWHw
• Also, when preprocessing your data, monitor both CPU and GPU usage and make sure you do not use unneccessary many resources. htop and nvtop are your friends! You can find some preinstalled CUDA environments under /opt/.

Important for CPU Users

• This Server is not meant for pure CPU usage. The servers CPUs are only meant for neccessary data preprocessing. Please contact the GWDG to get access to their High-Performance-Cluster if you run algorithms with high CPU load.