Remerciez-le!

Remerciez @Admin pour avoir partagé cet document gratuitement, de la manière la plus simple, en partageant sur les réseaux sociaux.

Bsp dev guide Parallel Computing Practical aspects BSPlib The BSP programming library Jonathan M D Hill a Bill McColl a Dan C Stefanescu b c Mark W Goudreau d Kevin Lang e Satish B Rao e Torsten Suel f Thanasis Tsantilas g Rob H Bisseling h a Oxford Unive

Parallel Computing Practical aspects BSPlib The BSP programming library Jonathan M D Hill a Bill McColl a Dan C Stefanescu b c Mark W Goudreau d Kevin Lang e Satish B Rao e Torsten Suel f Thanasis Tsantilas g Rob H Bisseling h a Oxford University Computing Laboratory Oxford OX QD UK b Harvard University Cambridge USA c Su olk University Boston USA d University of Central Florida Orlando USA e NEC Research Institute Princeton USA f Bell Laboratories Lucent Technologies N J USA g Columbia University New York USA h Utrecht University Utrecht The Netherlands Received October received in revised form April Abstract BSPlib is a small communications library for bulk synchronous parallel BSP programming which consists of only basic operations This paper presents the full de nition of BSPlib in C motivates the design of its basic operations and gives examples of their use The library enables programming in two distinct styles direct remote memory access DRMA using put or get operations and bulk synchronous message passing BSMP Currently implementations of BSPlib exist for a variety of modern architectures including massively parallel computers with distributed memory shared memory multiprocessors and networks of workstations BSPlib has been used in several scienti c and industrial applications this paper brie y describes applications in benchmarking Fast Fourier Transforms FFTs sorting and molecular dynamics Ó Elsevier Science B V All rights reserved Keywords Bulk synchronous parallel Parallel communications library One-sided communication Corresponding author E-mail jonathan hill comlab ox ac uk - see front matter Ó Elsevier Science B V All rights reserved PII S - - C J M D Hill et al Parallel Computing Introduction Since the earliest days of computing it has been clear that sooner or later sequential computing would be superseded by parallel computing This has not yet happened despite the availability of numerous parallel machines and the insatiable demand for increased computing power For parallel computing to become the normal form of computing we require a model which can play a similar role to the one that the von Neumann model has played in sequential computing The emergence of such a model would stimulate the development of a new parallel software industry and provide a clear focus for future hardware developments For a model to succeed in this role it must o er three fundamental properties Scalability the performance of software and hardware must be scalable from a single processor to several hundreds of processors Portability software must be able to run unchanged with high performance on any general purpose parallel architecture Predictability the performance of software on di erent architectures must be predictable in a straightforward way It should also ideally permit the correctness of parallel programs to be determined in a way which is not much more di cult than for sequential programs Recent research on Bulk Synchronous Parallel BSP algorithms architectures and languages has shown that the BSP model can achieve all of these requirements The BSP model decouples the two fundamental aspects of