Performance Analysis of WRF Model on Cluster and GPUs

Abstract

High Performance Computing is mostly used for Scientific Research. Various atmospheric models are used across the world, but the Weather Research and Forecast (WRF) Model is a fully functional modeling system for atmospheric research and operational weather prediction communities. If considered the weather models through which we achieve real-time forecasting and climate prediction, they are very time-critical and need a strong scaling. In this scenario only clusters with faster nodes and processors are not enough. With the introduction of CUDA cards in market, the computing power has raised its bar. In this paper, we have emphasized on the efficiency and scalability of WRF model on ParamVayu a HPC cluster at National Center for Medium Range Weather Forecasting (NCMRWF) and also on the performance scalability using the GPU cards. This research was done for benchmarking WRF model performance using various interconnects viz. PARAMNet-3, Infiniband and Gigabit. The paper also discusses on ways to increase the performance by changing node to core configuration and the GPU acceleration of WRF model using GPU Tesla c1060.