proach [1]. In this system, it is necessary to lay the knowledge about the impact of different approaches to or-
ganizing computations and working with memory on the final performance for different types of codes.
In this work, finite-difference 3D modeling of the propagation of elastic waves is considered [2]. Using the
example of solving this problem founded wide application, the influence of various optimization strategies for
parallel programs for various manycore architectures is investigated. Specific optimizations for different types
of architectures are considered, including improving the cache memory usage, balancing the computational
load, vectorization and accelerator memory usage.
Based on the research carried out, the software has been developed for various computing systems with
high rates of strong and weak scalability.
This research was supported by the Russian Foundation for Basic Research (grants No. 19-07-00085).
References
1. Glinskiy B.�., Zagorulko Yu.A., Zagorulko G.B., Kulikov I.M., Sapetina A.F., Titov P. A., Zhernyak G.F. Building
ontologies for solving compute-intensive problems // J. of Physics: Conference Series. 2021. V. 1715, Article Number
012071.
2. Sapetina A.F., Glinskiy B.�., Martynov V.N. Numerical modeling results for vibroseismic monitoring of volcanic
structures with different shape of the magma chamber // J. of Physics: Conference Series. 2021. V. 1715, Article Number
012057.
Solution approaches to numerical gas-dynamic problems with changing boundaries in LOGOS software
package
A. V. Sarazov, A. S. Kozelkov, D. K. Zelensky, R. N. Zhuchkov
FSUE �Russian Federal Nuclear Center � All-Russia Research Institute of Experimental Physics� Nizhny Novgorod
Region, Sarov
Email: avsarazov@vniief.ru
DOI 10.24412/cl-35065-2021-1-01-81
Approaches in simulation of the gas dynamic processes are of both scientific and practical interest. Transi-
ent modes in operation of engineering equipment that come from the motion of the element constituents of
the unit cause growing attention in the engineering practice. Numerical simulation of this class of problems in
LOGOS engineering software package [1] is available using two approaches different in the ideology: compu-
ting method on the deforming grids preserving links topology [2] and computing technology on overlapping
grids [3].
The choice of a particular approach comes from the immediate task setting. Nevertheless, there are some
problems where it is not possible to prefer either of the alternatives because it is impossible to describe com-
pletely the physical processes using one approach only.
The work reviews the realized physical-mathematical models for the problems of numerical gas dynamics
with moving structural components. It provides the examples of characteristic problems in aviation industry
that show operability of the realized models of the LOGOS software package.
References
1. M. A. Pogosjan, E. P. Savelevskikh, R. M. Shagaliev, A. S. Kozelkov, D. Yu. Strelets, A. A. Ryabov, A. V. Kornev, Yu. N.
Deryugin, V. F. Spiridonov, K. V. Tsiberev Application of Russian supercomputer technologies to develop the advanced
models of aviation technology VANT. Ser.: Mat. Mod. Fiz. Proc. 2013. No 2. P. 3-18.
2. E. Luke, E. Collins, E. Blades. A fast mesh deformation method using explicit interpolation, Journal of
Computational Physics 231 (2012) 586�601.
3. Deryugin Yu.N., Sarazov A.V., Zhuchkov R.N. Specific design features of the computing method on the Chimera-
type grids for non-structured grids// Mathematical simulation. 2017. Vol. 29, � 2. pp. 106-118.
Using Didal distributed data library for implementation of parallel fragmented programs for distributed
memory supercomputers
G. A. Schukin
Institute of Computational Mathematics and Mathematical Geophysics SB RAS
Novosibirsk State Technical University
Email: schukin@ssd.sscc.ru
DOI 10.24412/cl-35065-2021-1-01-82
Distributed data library Didal [1] is designed to facilitate development of efficient parallel programs for
distributed memory supercomputers. The aim of the library is to provide high-level distributed data structures
coupled with different strategies for data partitioning, distribution and load balancing, to develop parallel pro-
grams with. The library approach allows to use all existing tools for parallel program development, debugging
and profiling, as well as optimized C/C++ codes. One of defining features of Didal is its support of fragmented
programming [2]. In fragmented programming a parallel program is represented as sets of data and computa-
tional pieces (fragments), number and sizes of fragments being static or dynamic parameters. Results of im-
plementation of several parallel programs with Didal library and fragmented programming approach are pre-
sented. Details of parallel fragmented programs implementations, as well as key aspects of Didal's design, are
provided. The programs' performance is measured and compared with other parallel programming models.
References
1. Schukin G.A. Didal: distributed data management library for distributed memory supercomputers // Proceedings
of 13th international conference �Parallel Computational Technologies 2019 (PaVT 2019)�. 2019. P. 466.
2. Malyshkin V.E., Perepelkin V.A., Schukin G.A. Scalable distributed data allocation in LuNA fragmented
programming system // The J. of Supercomputing. 2017. V. 73, N. 2. P. 726-732.
3. Ivanov A. A. The title of the book. M.: Nauka, 1978.
Multi-grid starting initialization as a way to achieve higher convergence rates in industry-specific hypersonic
aerodynamic simulations
A. V. Struchkov, A. S. Kozelkov, D. K. Zelenskiy
FSUE �Russian Federal Nuclear Center � All-Russian Research Institute of Experimental Physics�, Sarov, Nizhny
Novgorod Region
Email: AnVStruchkov@vniief.ru
DOI 10.24412/cl-35065-2021-1-01-84
The paper presents a multilevel geometric initialization-based algorithm [1�2] to speed up external aero-
dynamics simulations. This method provides higher convergence rates and stability of numerical results in the
flow structure formation and settling phase. The concept of the method is to generate a series of coarse grids
[3] based on a parent grid to find solutions on each of the grids and to subsequently interpolate them to a fin-
er grid. The solution calculated on the finest of the grids constructed in series is then interpolated to the par-
ent grid, thus representing the starting initialization on it. The algorithm can be used on unstructured grids
with an arbitrary cell shape. As a cell combination criterion to form new control volumes in the successive grid
coarsening, the algorithm uses a relationship calculated from face areas and cell volumes. The cell combina-
tion process is based on the analysis of the weighted graph. Solution stability and convergence analysis was