Performance Characterisation and Evaluation of WRF Model on Cloud and HPC Architectures

Weather forecasting is a computationally intensiveproblem to handle as well as an important application problemwhere Cloud Computing can play a key transformational role.Traditionally the problem of handling such workloads havebeen considered mainly on High Performance Computing (HPC)architectures owing to their tightly-coupled nature and promisingspeed-up gains. However, with modern day Cloud platforms offeringa scalable service infrastructure, current HPC domain expertscontinue to deliberate on the usage of Cloud Computing platformsas an alternate cost effective and efficient way of procuring justin time computing. At the same time, there are a variety offactors to be considered before HPC workloads such as weatherforecasting can be moved to a Cloud infrastructure. In this paper,we have studied the performance characteristics of WeatherResearch and Forecasting (WRF) algorithm on three differentcomputing architectures. The architectures include two publicdomain Cloud service providers and a virtual machine (VM)based setup. Through detailed experiments, we have performeda rigorous comparative analysis of the performance of WRF ineach of these architectures and in that process we have identifiedthe strengths and weaknesses of each architecture for such HPCworkloads. We believe our work of comparing across 3 differentarchitectures is the first of its kind and could be seen as a pilotreference useful to domain experts and system administratorswho are considering to employ Cloud Computing for specificHPC problems.