The CompBioMed code HemeLB, developed at UCL was assessed by the POP Centre of Excellence to determine scaling ability of the code on SuperMUC-NG at LRZ. Peter Coveney and Jon McCullough at UCL worked with Brian Wylie at Jülich Supercomputing Centre to enable this assessment.
This audit examined strong scaling of HemeLB on the SuperMUC-NG Lenovo ThinkSystem with up to 309,696 MPI processes (on 6,452 compute nodes) using a 21.15 GiB testcase. Although the code always ran correctly, several compute nodes with notably inferior memory performance were identified and needed to be explicitly avoided. Compared to the smallest configuration that could be run using 864 processes (on 18 ‘fat’ compute nodes each with 768GiB), 190x speed-up was delivered with 80% scaling efficiency maintained to over 100,000 processes for the simulation phase.