Natural Load Indices (NLI) in NAMD2 load balancing algorithms

Existing measurement-based load indices generally provide adequate performance when used with dynamic load balancing algorithms in parallel scientific applications. An alternative that we present are natural load indices (NLI) that facilitate further performance improvement and better resource usage. Example NLIs are mass of an atom in a Molecular Dynamics (MD) code or rainfall amounts in a climate simulation. This paper presents performance results when we implement NLIs in NAMD2 a MD code. MD simulations are important for drug and medical research and require significant computational resources for long time periods. While an initial cost is incurred during code development in manually determining that NLIs exist, they can then be used at run-time with lower overhead and can reduce total execution time. Results indicate maximum improvement of 21% when comparing existing to NLI based algorithms with 10% overall improvement.