Multidimensional FFT on distributed-memory architectures for parallel computing
Tesi esterna in azienda
Parole chiave COMPUTATIONAL ELECTROMAGNETICS, ALGORITHMS
Riferimenti esterni M. A. Francavilla (Istituto Superiore Mario Boella)
Gruppi di ricerca Applied Electromagnetics
Tipo tesi MASTER THESIS
Descrizione When realistic complex geometries are considered, the numerical solution of Maxwell’s equations requires the development of ingenious fast and efficient algorithms. A possible approach is based on Fast Fourier Transform (FFT) for the computation of matrix vector products needed by an iterative solver(1). Porting this algorithm on distributed-memory architectures (cloud computing, grid computing) will dramatically expand the size of tractable geometries. This thesis aims at implementing a set of routines for the multidimensional FFT on distributed-memory architectures. The routines will be inserted in an existing numerical code and run on appropriate machines to face some of the most challenging problems in Computational Electromagnetics.
(1) F. Vipiana, M. A. Francavilla, G. Vecchi, ‘EFIE Modeling of High-Definition Multiscale Structures’, Antennas and Propagation, IEEE Transactions on , vol.58, no.7, pp.2362,2374, July 2010 doi: 10.1109/TAP.2010.2048855
Vedi anche http://www.ismb.it/en/node/2292
Conoscenze richieste - MS students in Computer Science
- Experience with main programming languages (Matlab /Fortran/C/C++)
- Knowledge of parallel computing (MPI)
Note Expected duration: 6 months.
The thesis will be developed at Istituto Superiore Mario Boella, Antenna and EMC Lab.
Scadenza validita proposta 16/01/2020 PROPONI LA TUA CANDIDATURA