HECToR - ARCHER Transition

Preliminary ARCHER access

This page will serve as a log of our activities and thoughts on processes and practices involved in making the transition from HECToR to ARCHER.

This is a first stab at listing some of the things which need to be done for the HECToR-ARCHER transition – to be fleshed out.

Maintain log of time/effort for the project.

  • Migration of data from /home and /work
    1. alert users to start doing this with some discernment
    2. check with/alert HECToR
    3. increase RDF quota to accommodate data
  • Advise users of possible differences in model results on ARCHER
    1. should they start new experiments yet
    2. can they finish work before HECToR switched off
    3. no chance of running on an XE6 after March
  • Decide which models we will support – take the opportunity to stop support for UM vn 7.1 for example – others? Should we only support certain configurations at specific resolutions?
  • Find out what access/data policies limit ARCHER usage (early access?)
  • What will be the equivalent of hum - should we set up a package account (like for oifs)?
  • What will be the new file system structure – hope it will be (insist that it is?) /work/n02/n02/… and /home/n02/n02/… otherwise lots of things could break
  • Set up a UMDIR - copy data

We need to make a test pair of installations for a set of standard configurations - one for Intel and one for Cray – possible conflict with namelist creation by small execs otherwise

How much of this can be done on the TDS?

Will the TDS have a serial queue?

  • HadGEM2-ES – extreme example
  • HadGEM3 model GA? Version?
  • NEMO (see what Andrew Coward is doing about NEMO)
  • Coupled job
  • UKV
  • Need pairs of GCOMs too

Testing these

  • Performance testing – compare Intel/Cray timings (Tom may have some useful info here?) – simple compiler options, we aren’t optimizing the models here.

  • Record results
  • We can use the MO’s Intel settings
  • Singapore Cray settings?

Model behaviour testing

  • Simon’s check_port for looking at model divergences?

  • should we get a scientist involved?
  • Record results

Install the UMs we wish to support

  • Back end and front end have same architecture
  • Shells/OS issues
  • Scheduler - script changes
  • Test small execs

  • Perform usual suite of porting tests

  • Consider other tests
  • Test other utilities xconv/xancil
  • No LMS machine - revamp archiving to use one of the post processors
  • CAP
  • Nesting
  • NWP tools NDDiag…


  • UMUI changes

  • Config files updated
  • Comms


Personnel - we will have extra resource devoted to Porting (inexperienced 'though)

Last modified 6 years ago Last modified on 01/07/15 14:34:24

Attachments (2)

Download all attachments as: .zip