Highly scalable dynamic load balancing in the atmospheric modeling system COSMO-SPECS+FD4

摘要: To study the complex interactions between cloud processes and atmosphere, several atmospheric models have been coupled with detailed spectral microphysics schemes. These schemes are computationally expensive, which limits their practical application. Additionally, our performance analysis of model system COSMO-SPECS (atmospheric Consortium for Small-scale Modeling SPECtral bin microphysicS) shows a significant load imbalance due to model. overcome this issue enable dynamic balancing, we propose separation scheme from static partitioning Using framework FD4 (Four-Dimensional Distributed Dynamic Data structures), show that approach successfully eliminates improves scalability system. We present balancing coupling two different supercomputers. The observed overhead is 6% on 1600 cores an SGI Altix 4700 less than 7% BlueGene/P at 64Ki cores.