Skip to content

Latest commit

 

History

History
154 lines (115 loc) · 5.6 KB

File metadata and controls

154 lines (115 loc) · 5.6 KB

AFS

WebAFS: https://afs.stanford.edu/ Documentation: https://uit.stanford.edu/service/afs

Clusters

Farmshare

Login nodes: rice01 ... rice15 (but not all numbers in that range)

  • List all rice login nodes:
    dig +noedns +norecurse +short +vc @lbdns1.stanford.edu rice.best.stanford.edu TXT | awk '{ print $3 }' | sed -e "s/\"//g"
    

Notes

  • AFS access
    • Only rice nodes have AFS access, mounted at /afs
      • ~/afs-home conveniently links to users' AFS home directories
    • To check quota and usage: fs listquota ~/afs-home

Documentation

Sherlock

Login nodes: sh-ln01 ... sh-ln08

OnDemand: http://sherlock.stanford.edu

  • Accessing open ports on nodes: https://login.sherlock.stanford.edu/rnode/<node>.int/<port>/
    • Does not seem to work well with Jupyter notebooks

Provided utilities: sh_...

  • Type sh_ and then press <TAB> twice to see all available utilities.

Documentation: https://www.sherlock.stanford.edu/

SCG4: Stanford Genomics Cluster

Login nodes: scg4-ln01 (smsx10srw-srcf-d15-35) ... scg4-ln04 (smsx10srw-srcf-d15-38)

Documentation: https://web.stanford.edu/group/scgpm/cgi-bin/informatics/wiki/index.php/Main_Page

Interactive Environments

Jupyter

jupyter notebook --no-browser --ip=0.0.0.0 --port=<port>

  • --no-browser: do not automatically search for installed browsers in which to open the notebook after startup
  • --ip=0.0.0.0: listen to all incoming ports, not just those from 127.0.0.1 (localhost) on the remote machine itself
  • --port=<port>: port that the notebook server will listen on

Farmshare

  • Jupyter needs to be installed by the user, e.g., in a conda environment.
  • On compute nodes (wheat, oat), run the following lines before launching Jupyter, and choose a sufficiently high port number > 32768 that is not firewalled
    mkdir /tmp/$SLURM_JOBID
    export XDG_RUNTIME_DIR=/tmp/$SLURM_JOBID
    jupyter notebook --no-browser --ip=0.0.0.0 --port=<port>
    

Sherlock

  • Jupyter is available as a Lmod module
    module load py-jupyter/1.0.0_py36
    module load py-scipystack/1.0_py36
    
  • Provided script: sh_notebook
    • Use -h option to see usage documentation
    • Runs Jupyter notebook as a job on a compute/dev node
    • Loads default Python version (2.7.13, as of Feb. 2019)

RStudio

Farmshare, Sherlock

module load rstudio
module load R/3.5.1 [Sherlock only]
rstudio

This requires remote display, e.g., X11 or VNC (Farmshare only).

RStudio Server

rserver --www-address=0.0.0.0 --www-port=<remote_port> --auth-validate-users=1

  • --www-address and --www-port are equivalent to --ip and --port for jupyter notebook
  • --auth-validate-users=1: validates that authenticated users exist on the host system

Farmshare: not available (see Unanswered Questions below)

Sherlock

  • RStudio Server is available as part of the rstudio Lmod module. By default, loads R 3.4.0 (as of Feb. 2019), unless the desired R module is loaded after loading rstudio as shown below:
    module load rstudio
    module load R/3.5.1
    
  • Provided script: sh_rstudio
    • Use -h option to see usage documentation
    • Runs RStudio Server as a job on a compute/dev node
    • Loads default R version (3.4.0, as of Feb. 2019)

SSH Tunneling

Host Direct forward Forward through login node Hop-forward through login node
Farmshare (login node) yes n/a n/a
Farmshare (compute node) yes yes no*
Sherlock (login node) yes n/a n/a
Sherlock (compute node) yes yes yes
*due to two-factor authentication between login and compute nodes

Direct forwarding: ssh -N -L localhost:<loc_port>:localhost:<host_port> <user>@<remote>

  • Directly to host: set <host> = localhost, <remote> = hostname of host
  • Through login node: set <host> = hostname of host, <remote> = hostname of login node

Multiple hop port forwarding: ssh -L localhost:<loc_port>:localhost:<middle_port> <user>@<middle_host> -N -L <middle_port>:localhost:<remote_port> <remote>

See linux.md file for more information.

forward utility

Questions

Unanswered

Sherlock

  • How to install R via conda on Sherlock?
  • How to load additional packages or activate pip/conda environments on Sherlock when using the provided sh_dev/sh_notebook/sh_rstudio scripts?

Farmshare

  • How to run RStudio Server on Farmshare?

Myth

  • How to create pip or conda environments on Myth?

Answered

General