A comparative analysis of the performance and deployment overhead ofparallelized Finite Difference Time Domain (FDTD) algorithms on a selection of high performance multiprocessor computing systems

[摘要] ENGLISH ABSTRACT: The parallel FDTD method as used in computational electromagnetics is implemented on a variety ofdifferent high performance computing platforms. These parallel FDTD implementations haveregularly been compared in terms of performance or purchase cost, but very little systematicconsideration has been given to how much effort has been used to create the parallel FDTD for aspecific computing architecture. The deployment effort for these platforms has changeddramatically with time, the deployment time span used to create FDTD implementations in 1980ranging from months, to the contemporary scenario where parallel FDTD methods can beimplemented on a supercomputer in a matter of hours. This thesis compares the effort required todeploy the parallel FDTD on selected computing platforms from the constituents that make up thedeployment effort, such as coding complexity and time of coding. It uses the deployment andperformance of the serial FDTD method on a single personal computer as a benchmark andexamines the deployments of the parallel FDTD using different parallelisation techniques. TheseFDTD deployments are then analysed and compared against one another in order to determine thecommon characteristics between the FDTD implementations on various computing platforms withdiffering parallelisation techniques. Although subjective in some instances, these characteristics arequantified and compared in tabular form, by using the research information created by the parallelFDTD implementations. The deployment effort is of interest to scientists and engineers consideringthe creation or purchase of an FDTD-like solution on a high performance computing platform.Although the FDTD method has been considered to be a brute force approach to solvingcomputational electromagnetic problems in the past, this was very probably a factor of the relativelyweak computing platforms which took very long periods to process small model sizes. This thesis willdescribe the current implementations of the parallel FDTD method, made up of a combination ofseveral techniques. These techniques can be easily deployed in a relatively quick time frame oncomputing architectures ranging from IBM's Bluegene/P to the amalgamation of multicore processorand graphics processing unit, known as an accelerated processing unit.

[发布机构] Stellenbosch University