Getting Started using the PPPL cluster

To use the PPPL cluster, do the following:
  1. if you don't already have one, get a PPPL Unix account by contacting Lena Scimeca in Computer Services
  2. log into portal.pppl.gov, the login "front end" of the cluster, using ssh (secure shell)
  3. copy or compile your program so it is in your home directory or one of its subdirectories
  4. create a script to run your program
  5. submit that script to the batch job scheduler for execution in a batch queue
The batch job scheduler will check that the attributes you have specified in your job are met, and that there are enough hosts available. If all conditions are met, the scheduler starts the job.

Typically, a job contains programs that are parallelized, whose threads each run on individual CPUs, and which run on more than one host simultaneously.

Here's an example:
You have a multi-threaded, parallelized program, and have compiled this program using the appropriate parallel compiler. You want to use a facility called mpirun to run the program on multiple hosts. You would:
  1. create a script that will run your program (see sample script)
  2. submit this script to the batch job scheduler (see submitting your job)
  3. collect your output
Interactive use
The PPPL cluster also has two types of interactive hosts
  1. Non-exclusive: A set of hosts has been provided with the alias portal.pppl.gov. When you login to portal.pppl.gov, logins are allocated in a least loaded fashion to one of the hosts in this group. Here, you can do compilations, do test runs, or submit jobs to the batch queue. The portal hosts are shared by all users, so running very high CPU or memory intensive jobs is discouraged. In that case, you should reserve a system for your exclusive use as detailed in the next section.
  2. Exclusive: You can reserve one or more hosts for your exclusive use (i.e. no other users will be able to use the hosts allocated to you) by ssh'ing to portal.pppl.gov and typing the command 'use'. This command will ask you a set of questions, like the number of hosts you need and if you require a host or hosts with a large memory size. It will then reserve a set of hosts on which you can run programs interactively. The advantage of using this approach is that the job scheduler will reserve your hosts for your use only and prevent other users from running on the hosts you have been allocated.