Adun Cluster Usage Tutorial

The current procedure to use ADUN on the cluster is the following:

Create the simulation:

  • Use the GUI in order to create your systems, templates and the simulation.
  • Export the simulation to a folder which is accessible by the cluster, i.e. /cbbl/users/YOUR_NAME or /cbbl/users/scratch. So the full path to your simulation should look like this: /cbbl/users/YOUR_NAME/SIM_DIR

Tell the cluster something about your plans

If you would like to have a bit more control about what the cluster is doing, you can create a file in the simulation directory that includes some settings. This, however is not necessary. If you don’t supply a settings file, default values will be used. Go to your simulation directory /cbbl/users/YOUR_NAME/SIM_DIR. Create a file cluster.ini, open it and put in some of the following things:

name=STRING

  • The name must not begin with a number and must not contain whitespaces
  • default: the name of your simulation directory

queue=unlimited / cbbl / short-vitis / short-pc

  • Please inform yourself on http://moebius.imim.es/grib_resources/ there you’ll get a description on which queue to use when
  • default: unlimited

duration=HH:MM:SS

  • Any sensible time. HECTOR: max 12h LAPALMA: max 72h

template=NAME

  • Has to be the name of your template file
  • default: template.tem

tasks=NUMBER

  • The amount of tasks you need to start E.g. For a RepEx simulation with 8 replics, put here a 8

threads=NUMBER

  • This differs from supercomputer to supercomputer. On Hector, this specifies, how many cores in a node you’d like to allocate for MPI processes. For Hector, you put here 4 for the XT4 and 24 for the XT6.
  • For Lapalma, this specifies how many cores will be used for one thread. Put here a 1 if you’re not doing extra therading. For extra threading, put a 2

tasksPerNode=NUMBER

  • This specifies How many mpi processes should be executed on each node. Without extra threading put here 1

initialMinimisation = YES / NO

  • want an inital minimisation?

system = pbs / slurm / sun

  • For the CBBL-queue use sun, for Hector use pbs and for LaPalma use slurm

Push the job onto the cluster:

Now, at the moment you need a linux box to start the simulation, though this will change soon.

  • Go to /cbbl/soft/adun/production
  • Push a job onto the cluster by typing ./cluster PATH_TO_SIM_DIR e.g. ./cluster /cbbl/users/YOUR_NAME/SIM_DIR

voilà, if you did everything correctly, you will see the list of previously submitted jobs. If not, there’ll be some errormessage

Leave a Comment