Title: Peculiarities of numerical algorithms parallel implementation for exa-flops multicomputers
Authors: Victor E. Malyshkin
Addresses: Supercomputer Software Department, Institute of Computational Mathematics and Mathematical Geophysics, Russian Academy of Sciences, 6, Pr. Lavrentieva, 630090, Novosibirsk, Russia
Abstract: Peculiarities of numerical algorithms parallel implementation for exa-flops multicomputers were considered and the appropriate examples were given. The problems of such big data processing were analysed and also, a solution was suggested. Problems of parallel implementation of large scale numerical models on a rectangular mesh were demonstrated by the parallelisation of the particle-in-cell method (PIC). For similar problems solution, the fragmentation of big data and computations were suggested. Fragmentation automatically provides different useful properties of a target program including dynamic load balancing on the basis of fragments migration from overloaded into underloaded processor elements of a multicomputer.
Keywords: numerical algorithms; big data processing; parallel programming; large scale numerical models; exa-flops computation; fragmentation; fragmented programs; dynamic load balancing; processes migration; modelling; particle-in-cell; PIC; multicomputers.
DOI: 10.1504/IJBDI.2014.063837
International Journal of Big Data Intelligence, 2014 Vol.1 No.1/2, pp.65 - 73
Received: 01 Oct 2013
Accepted: 03 Feb 2014
Published online: 30 Sep 2014 *