Abstract
The standard method used in the Weather Research and Forecasting (WRF) model for distributing MPI processes across the processors is not always optimal. This circumstance affects performance, i.e., execution times, but also energy consumption, especially if the application is to be extended to exascale. The authors found that the reason why the standard method for process distribution is not always optimal was an imbalance between the orthogonality of the communication and the proper cache usage, and this affects energy consumption. We present an improved MPI process distribution algorithm that increases the performance. Furthermore, scalability analyses for the new algorithm are presented and the energy use of the system is evaluated. A solution for balancing energy use with performance is also proposed for cases where the former is a concern.
Highlights
Weather forecasting is becoming increasingly important in people’s everyday lives
From an operational point of view, performance is the most important computational aspect of systems providing weather forecasts, as these must be generated in a short period of time, without losing sight of the energy consumption of these computational resources. us, a forecast for the 12 hours should be computed in less than an hour in order to be useful for operations
Increasing the number of processors used on a weather simulation allows the problem to be split into smaller subproblems, but at the cost of an increased communication throughput. is increase in computational resources cannot be sustained indefinitely; at some point, the computational workload of the subproblems will be so low that communications will become a bottleneck, avoiding further reduction in the computation times
Summary
Weather forecasting is becoming increasingly important in people’s everyday lives. It is as important for a person who wants to have a good weekend as it is for an agency that has to plan a world-class event like the Winter Olympics. There was an enormous difference in performance between the two layouts For this reason, we studied the impact of the different process distributions on the simulation times and the reasons for this impact and proposed a new distribution algorithm that works better than the one implemented by WRF. We studied how increasing the number of used processing resources decreases the wall time of the simulation at the cost of losing efficiency for computing the same workload, dramatically increasing the energy consumption.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
