Summary¶

Today we have discussed:

• Basic architecture of a typical HPC cluster

  • Nodes as building blocks
  • Login nodes - don’t do heavy lifting here
  • Compute node
  • GPU nodes

• Job scheduler concepts

  • You interact with the scheduler on the front end
  • The scheduler moves your “run” to the compute nodes
  • Waiting queue

• Job script

  • Description and requirements of your job
  • Request resources, e.g.:
    • Job time
    • Number of nodes/cores
    • Amount of memory
    • Number of GPUs
  • Includes UNIX script to describe your job to the system
    • Program(s) to execute
    • Location of input data
    • Where to put the output data
    • Disks to be utilised
    • You can be creative here ;)