Weather Forecast using Cloud Computing with SyncHPC

Numerical weather prediction (NWP) system is designed to simulate and predict the atmosphere with a computer model. Weather Research and Forecasting (WRF) is a set of software applications used for this prediction. WRF is used extensively for research and real-time weather forecasting around the world.

WRF application needs High Performance Computing (HPC) clusters connected using high speed interconnect like InfiniBand for better run-time duration. Syncious team demonstrated WRF application performance on HPC cluster deployed in cloud using SyncHPC platform. This blog discusses the results of this run.

HPC Cluster Details:
24 compute nodes (384 CPU Cores).
Interconnect: Infiniband (FDR speed: 56Gbps)
OS : CentOS 7.4
CPU : Intel Xeon E5-2667
MPI : Intel MPI

Commands used to run WRF simulation:
Using WRF Preprocessing system (WPS) below commands created input files for WRF run.
./ungrib.exe
./metgrid.exe

Above commands prepared input to the real program for real-data simulations which were run using the commands mentioned below.

$MPI_DIR/bin/mpirun –np <number of cpu cores used> ./real.exe
$MPI_DIR/bin/mpirun –np <number of cpu cores used> ./wrf.exe

The forecast duration was kept constant for all the runs with different numbers of CPU cores (16, 32,64, 128, 356, 384). Time required for execution of WRF programs was used as a metric to measure performance.

Performance improvement is almost linear on scale as the number of CPU cores are increased.

wrf_scale

Syncious team observed almost linear reduction in time required for WRF simulation with increase in number of CPU cores.

wrf_time

Results show that WRF can be run on SyncHPC cloud as efficiently as it would be using on-premise HPC cluster. Additionally, SyncHPC platform provides multiple features to use and manage cloud-based HPC resources efficiently.