Authors
Pomelov S.
|
Degree
|
Postgraduate, Institute of Computational Mathematics and Mathematical Geophysics SB RAS |
|---|---|
|
E-mail
|
s.v.pomelov@gmail.com |
|
Location
|
Novosibirsk |
|
Articles
|
Parallel programming: simulation of the algorithm execution on a given architectureAn important part of modern science and economics are numerical computations in mathematical
models. Their efficiency depends on the effectiveness of the hardware. This, in turn, is related to
parallel computations’ optimizations. A new approach to the optimizations and developing a tool for
parallel computing process optimization are the goals of the ongoing research. These software tool
development principles and other resulted applications are described in the paper. Properties of computer
architecture are suitable for formalization. Parallel algorithms already exist in a mathematical
form. From the two points, two mathematical models are derived: the model of the computational algorithm
and the computer architecture model. Along with other points, here described their properties
and applications. Both models, even though they represent vastly different real-world objects,
are described in the same terms and with the equivalent level of detail and structure. This distinction
from similar works allows developing a new approach to comparative analysis of the algorithm scalability,
as well as algorithm efficiency for a given architecture. These models are used as a basis for
a core module of the software implementation for simulation and optimizing the architecture and algorithm
interactions. In this article, after the computational process simulation description with the
models’ usage, as a midline working result, placed suggestions for practical using the software, proposed
approach and results of those simulations. At the ending, a few practical examples are demonstrated.
The last simulation example is crosschecked with a computational experiment, derived important
conclusions from its results.
Read more...
|
