There are couple of methods available to improve performance of Star-CCM+ jobs in parallel environment using IBM Platform MPI on HPC.
Here, I discuss about Core locking / CPU binding method which significantly improves performance. Core locking will binds a rank to a logical processor (cores) to prevent a process from moving to a
different logical processor after start-up.
Core locking can be done using “-aff ” mpi option. Below is the syntax with core locking option in Star-CCM+ version 8.06
starccm+ -rsh ssh -mpidriver platform:”-prot -IBV -aff=automatic:default:core” -batchsystem <lsf /pbs /sge> -power -batch -np <no of cores> <input file>
Detailed options can be found in IBM Platform MPI User’s Guide.
Here is the benchmark result before & after the Core locking implementation done.
Benckmark Case: HL-20 MACH 10 (Hypersonic Drag Polar – 6.4 million cell single region polyhedral mesh – Coupled solver)
|S. No||No of cores||~Time taken / Iteration (in sec)|