In this paper we report computational benchmarks and scaling results of finite difference time domain solutions of Maxwell's equations for high intensity RF interactions with aircraft using both a local high performance computational cluster and a cluster created out of on-demand commercial systems available from the Amazon Elastic Computer Cloud. We observe good scaling and comparable performance to local HPC clusters. This result implies strong potential for using cloud resources for FDTD problems of interest to electromagnetic effects teams. I. INTRODUCTIONThe finite difference time domain (FDTD) method of solving Maxwell's equations using the Yee scheme has been valued because of its practical use and elegant simplicity. It is the latter attribute that has greatly expanded its application in recent years due to its great amenability to parallelization [1].Advances in the generation and maintenance of FDTD models from detailed CAD drawings combined with dramatic increases in available computing resources and parallel algorithms make FDTD a compelling choice for many classes of electromagnetic effects (E3) programs interested in modelling large systems and air/space craft down to individual electronics interfaces. However, full aircraft high intensity RF (HIRF) simulation at frequencies above 3 GHz and lightning simulations with computational cells sizes below 10 mm are tantalizingly close to being economical with current computer systems [2]. This paper is motivated by the desire to study the potential of cloud HPC resources to obtain advanced computational cluster resources without the cost and burden of maintaining a full-time, dedicated HPC center.As computational resources and data center efficiencies increase over time, the numerical techniques that scientists and engineers employ to take advantage of them must evolve as well. As single-processor speeds were the primary driver of improvement in throughput, computational scientists needed to only focus on the efficiency of their code in singleprocessor operation. In recent years, much of the computational throughput has come through an increase in the number of computational cores on a single chip. Further, increases in the bandwidth of interconnects between individual compute nodes makes certain classes of electromagnetic simulation problems computationally accessible through the use of high performance computing (HPC) clusters of several individual compute units. This paradigm has dominated the landscape for organizations and

No abstract

No abstract

No abstract