An Integrated Topology-Aware Mapping Scheme for Large-Scale CFD Applications on the ORISE Supercomputer
摘要
The diversity in network topologies and internal node architectures is driving the emerging trend of hierarchical process mapping studies. In this paper, we apply a integrated process mapping scheme aimed at enhancing communication performance for large-scale CFD applications on the ORISE supercomputer. This scheme includes a top-down domain decomposition method, utilizes network topology for inter-node process mapping, and explores two optimization strategies for intra-node process mapping. Experimental results show that this scheme can effectively improve the communication performance of two CFD applications (OpenCFD and CCFD V3.0) on the ORISE supercomputer.