Skip to content

Latest commit

 

History

History
80 lines (66 loc) · 2.46 KB

DAS5_cheatsheet.md

File metadata and controls

80 lines (66 loc) · 2.46 KB

DAS-5 Cheat Sheet

  • In the first LC (02-03-2020) you will be assigned account names for DAS-5, a computing cluster that you will be expected to run some experiments on.
  • DAS5 uses a scheduler called SLURM, the basic procedure to use it is first reserving a node (or a number of them) and then running your script on that node.
  • You are recommended to use the VU node of DAS-5 (fs0.das5.cs.vu.nl)
  • You are not allowed to use the account name given to you for this course for any other work such as training NN etc...
  • For information about DAS-5 see https://www.cs.vu.nl/das5/

DAS-5 Usage Policy

  • Once you log in to DAS-5 you will be automatically in a shell environment on the headnode.
  • While on the headnode you are only supposed to edit files and compile your code and not execute code on the head node.
  • You need to use prun to run your code on a DAS-5 node.
  • You should not execute on a DAS-5 node for more than 15 minutes at a time.

Connecting to DAS-5

Use ssh to connect to the cluster:

ssh -Y [email protected]

Enter your password, If its correct then you should be logged in.

If you are a MacOS or Linux user, ssh is already available to you in the terminal.

If you are a Windows user, you need to use a ssh client for Windows. The easiest option is to use putty: http://www.chiark.greenend.org.uk/~sgtatham/putty/download.html

Commands

Check which modules are loaded at the moment

module list

Check which modules are available

module avail

Load prun

module load prun

Load cuda module

module load cuda80/toolkit

Show nodes informations ( available hardware etc )

sinfo -o "%40N %40f"

Show active and pending reservations

preserve -llist

Reserve a node with the ’cpunode’ flag for 15 minutes:

  • BE AWARE: 15 minutes is the maximum time slot you should reserve!
  • In the output of your command you will see your reservation number
preserve -np 1 -native '-c cpunode' -t 15:00

Run code on your reserved node

prun -np 1 -reserve <reservation_id> cat /proc/cpuinfo

Get a randomly assigned node and run code on it

prun -np 1 cat /proc/cpuinfo

Schedule a reservation for a node with a gtx1080ti gpu on christmas day ( 25th of december )starting at midnight and 15 minutes for 15 minutes

preserve -np 1 -native ’-c gpunode,gtx1080ti’ -s 12-25-00:15 -t 15:00

Cancel Reservation

preserve -c <reservation_id>