Flux allocation for Stats 531

Edward Ionides

  • The folks at Flux have kindly set up a class Flux allocation for Stats 531.

  • Using Flux is entirely optional. If you’re new to Flux, but prepared to work in a Unix environment, then the online tutorial “Flux in 10 easy steps” may indeed be not too difficult to follow, and a worthwhile investment of time.

  • Flux has excellent support staff, if you have questions, but you’re welcome to start by asking them to Dao or myself.

  • Technical instructions about the Stats 531 Flux allocation follow:

  • A 100 core allocation has been created for stats531w16_flux.

  • These cores are available now and will remain active for 1 month, through the end of the day on 5/14/2016.

  • Everyone on the Canvas email list for Stats 531 has been added as authorized users.

  • For students to use this allocation, they should include the following lines into their PBS scripts:

    #PBS -A stats531w16_flux
#PBS -l qos=flux
#PBS -q flux

  • The available resources can be checked quickly with the freealloc command on Flux:

    freealloc stats531w16_flux
0 of 100 cores in use, 100 cores available
0 KB of 400 GB memory in use, 400 GB memory available

  • An example of doParallel use is in the example files in /scratch/data/examples/R/doParallel. Type the commands

    $ mkdir example
$ cd example
$ cp /scratch/data/examples/R/doParallel/*  ./

    to make local copies of the following files:

    doParallel_example.pbs
doParallel_example.R
doParallel_example.Rout
test_R_doParallel.o19081388

  • You have to make a few modifications to run your code.

    • In the PBS submit script, there are instructions in the comments at the head of the file. You need to fill in your e-mail address and Flux account name before submitting it.

    • In the R program itself, there is an example of a parallel foreach operation. This should be replaceable by the foreach code we’ve been using for POMP analysis.