In order to efficiently utilize the computational resources at hand, a short benchmarking study was run. Two systems of varying sizes were chosen and simulated using two different MD suites on two Opteron clusters with different interconnects.
The resulting efficiency plots are shown in Figure 1 along with the simulation details. In all cases, the use of Infiniband interconnects effectively extends the range of processors over which simulation can proceed efficiently. NAMD was seen to scale more efficiently for larger systems ( > 100,000 atoms) at higher processor numbers (> 64 processors), while for smaller systems the scaling of the two programs was approximately equal. Setting an acceptable loss in efficiency limit of 60% our large-scale biomolecular simulations should be run on 64 processors. On this number of processors there is little difference in scaling between the two MD suites. However, for reasons of lipid compatibility, NAMD is preferred.
Figure 1: Simulation efficiency over a range of processors numbers. Systems: DHFR (31,000 atoms) and CDK2 (109,000 atoms). MD suites: AMBER9 and NAMD 2.6. Interconnects: Infiniband and Gigabit Ethernet.
Full details of this benchmark study can be found in Benchmark Report
We would like to acknowledge the Institute for Information Technology and Advanced Computation (IITAC), the Higher Learning Authority (HEA) PRTLI scheme, the National Development Plan (NDP), the Trinity Centre for High Performance Computing (TCHPC) and the SFI/HEA Irish Centre for High-End Computing (ICHEC) for the provision of computational facilities and support.