Computational fluid dynamics on the AWS cloud

Introduction

Warnings

Procedure

  • Resource tagging per HPC Club policy
    • Link
    • Project tag key; value = NetID (without ‘@uw.edu’)
    • End_date tag key; value = when you expect to be done with the resource
    • Name tag key; value = NetID_descriptive_phrase

  • Central idea: Use the Spot market, do the first calculation, make sure it works, get cost

  • AB can assist with getting the technical details right

  • Authenticate with NetID possible but we will use IAM User for now

  • Procedural
    • Create cfn cluster with placement group
      • See email chain for links
    • optional: disable hyperthreading (config file and post_install_script required)
    • install starccm+ on the shared volume, set up the license
      • A making progress on the execution issue; please send Rob confirmation this is fixed
    • create a snapshot of the volume for next cluster
    • create a qsub file and launch
      • created from starccm+ documentation; get file from Siemmens, cf email chain; pls send to Rob
  • Number of processes / instance depends on number of gridcells in the CFD calc
    • LH: 50k cells per process
    • LH: 72 processes implies qty 2 c4.8xlarge with hyperthreading on; qty 4 if off
    • A revises: 9e6 cells 180 processors = 4-5 c4.8xl
    • ‘existing config does not have a placement group’ implies new cfncluster needed
  • Obstacle
    • Insufficient space on the shared EFS drive… solution pending
  • The Steve’s Portal
    • This is the starccm+ website
    • batch/qsub documentation available
Tags: research_computing scale