Novosibirsk, Russia, May, 30 – June, 4, 2011

International Conference
"Modern Problems of Applied Mathematics and Mechanics: Theory, Experiment and Applications", devoted to the 90th anniversary of professor Nikolai N. Yanenko

Sapronov I.   Быков А.Н.   Воронин Б.Л.   Ерофеев А.М.  

RAMZES-KP Technique

Reporter: Sapronov I.

RAMZES-KP technique [1] is meant for computation of multicomponent heat-conducting media motion in Euler-Lagrange coordinates using distributed memory parallel computational systems.
The technique is based on the following principles:
• physical processes and space directions splitting;
• use of gas dynamics and heat conduction equations in Cartesian and curvilinear coordinate systems, as well as in Euler-Lagrange variables;
• use of implicit finite-difference time approximation as heat conduction equation, as well as gas dynamics equations;
• decomposition of problem geometry into fragments interacting through boundary conditions transfer;
• use of lumping and Yangs methods to reconstruct interfaces in computing matter flows from mixed cells.
Fractional step method developed by N.N. Yanenko is widely used in RAMZES-KP technique.
The specific feature of this technique is the use of parallel computations at every stage (preliminary stage, computation and results analysis) of problem run on multiprocessor distributed memory computer.
Non-reconstructible sub-matrix decomposition of data matrix was used when developing parallelization methods within the time step. In addition to advantages of this technique, there are difficulties connected with run parallelization. One of the first works on run “parallelization” was published in 1978 by N.N. Yanenko [2].
The own version of parallel-pipeline method has been developed for RAMZES-KP run parallelization. The peculiarities of the suggested implementation (automatic control of pipeline piece number, independence of cross runs, independence of grid sheet runs, etc.) are described in the report. The developed parallelization methods allowed the use of modern high-performance multiprocessor computers (50-60% for the time step of several seconds, the performance increases up to 60-80% for a longer time step).

1. Bykov A.N., Veselov R.A., Voronin B.L., Erofeev А.М. RAMZES-KP Technique for Computation of Multicomponent Heat-Conducting Space Motions in Euler-Lagrange Coordinates. // RFNC-VNIIEF Proceedings. 2008. Isue 13. (in Russian)
2. Yanenko N.N., Konovalov A.N., Shustov G.V. On Organization of Parallel Computations and Run “Parallelization”. – Rep. Numerical Methods of Continuum Mechanics. Novosibirsk, 1978, V.9, N 7, Pp.139-146. (in Russian)

Abstracts file: Sapronov-e.doc

To reports list
© 1996-2019, Institute of computational technologies of SB RAS, Novosibirsk