General Information

Top of Page

User Guide for BNL Blue Gene/Q


Overview

This user guide documents topics of interest to the end users of the BNL Blue Gene/Q such as using the compilers and job submission.

It is an IBM pre-production machine and currently not as user-friendly as NY Blue/L and NY Blue/P which are production machines, e.g. currently the IBM compilers and the IBM ESSL library (optimized for Blue Gene/Q) are not available though we are attempting to obtain them. Also at the moment there is no batch system.

Local and some offsite documentation can be found in the sidebar at the left.

The hostname for the front end node of the BNL Blue Gene/Q is fenq.qcdoc.bnl.gov, see the Logging In page.

Blue Gene is not a typical parallel computer. It is meant for codes that scale well into many thousands of processors.

The compute nodes on which your job runs are rebooted prior to running your job on them, ensuring that there is no "junk" remaining from whatever jobs ran on those nodes previously. The nodes are also exclusively allocated to your job.

If you find any errors on these pages please notify BNL Blue Gene/Q Web Administrator

Top of Page

Simple Example: Compile, Link, Run

  1. Compile and link file hello.c which contains a C language MPI program:
    mpixlc_r -o helloc hello.c
     
  2. Submit the job interactively:
    runjob --block BLOCKID --exe /home/johndoe/helloc -p 16 --np 2048 --env-all --cwd /home/johndoe > job.output 2>&1

Notes pertaining to this Example

If for some reason mpixlc_r is not in your path, see the BNL Blue Gene/Q Compiler Invocation page to see how it and all the compilers and their mpi wrappers are invoked.

BLOCKID is a blockid you are given when you are approved for a BNL Blue Gene/Q account.

Currently 128 node and 512 node blocks are being granted, and 512 node blocks are more difficult to obtain. The BNL Blue Gene/Q only has 1 rack (1024 nodes).

This example runs 16 mpi tasks per compute node and uses 2048 mpi tasks for the entire job, and it assumes you are using a 128-node block. There are 16 cores per compute node available for your job on the machine, hence this example uses every one of them.

Please study the following man page: man runjob

Note that two preceding hyphens are specified for all the flags used in this example except for -p, for which only one preceding hyphen must be specified, as detailed on the man page cited above.