RZ Bull

Hardware

• 27,880 total cores and huge but undetermined amount of total RAM installed in October 2011:
  • 1,358 Bullx Blade B500 @ 12 3.06 GHz Xeon X5675 Westmere EP
  • 104 Bullx S6030 @ 32-128 2.00 GHz Xeon X7550 Nehalem EX
• A quick comparison for the dallas.0153 test case on 8 PEs shows 2500 ts/hr for the AICES Xeon cluster, 5600 ts/hr for the RZ Bull SMP Nehalem nodes, and 7100 ts/hr for the MPI Westmere nodes.

Links

• Rechenzentrum: primer • how-to • wiki • configuration • mailing list

LSF Batch Job System

Sample LSF MPI job file, say run.j, looks like this:

#!/usr/bin/env zsh

#BSUB -J test      # job name
#BSUB -oo run.qout # job output (use %J for job ID)
#BSUB -We 5 -W 10  # soft and hard limits in hours:minutes
#BSUB -M 512       # memory in MB
#BSUB -n 8         # number of slots
#BSUB -a openmpi   # MPI environment
#BSUB -m mpi-l     # host group (mpi-s or mpi-l for Westmere, smp-s or smp-l for Nehalem)

$MPIEXEC $FLAGS_MPI_BATCH $HOME/bin/i386/xns

Submit it with bsub < run.j. Look at the status with bjobs. Kill all own jobs with bkill 0.

Parallel environments