|
|
 |
• Содержание выпуска •
• Mathematical Foundations of Programming •
• Software and Hardware for Distributed Systems and Supercomputers •
• Artificial Intelligence, Intelligence Systems, Neural Networks •
• Supercomputing Software and Hardware •
• Software and Hardware for Distributed Systems and Supercomputers •
Software and Hardware for Distributed Systems and Supercomputers
Responsible for the Section: Sergei Abramov, Dr. Phys.-Math.Sci.,
corresponding member of RAS
On the left: assigned number of the paper, submission date, the number
of A5 pages contained in the paper,
and the reference to the full-text PDF
/r1/pdf.jpg) .
|
Article # 2_2017
16
p.
/r1/pdf.jpg)
PDF |
submitted on 28th
Dec 2016 displayed on
website on 23th Jan
2017 Dmitri Knyazkov
Effective computation of two-dimensional FFT on a homogeneous or
heterogeneous cluster
The paper considers performing two-dimensional FFT on
a supercomputer. It investigates a dependance of FFT computation
time from a matrix size for MVS-100K, MVS-10P and HybriLIT
supercomputers. A method of CPU-GPU load balance for a heterogeneous
cluster is proposed. For a TESLA K40 card it is shown, that
two-dimensional FFT computation time is almost equal to data
transferring time. The computation itself is 48 times faster when
using GPU comparing to two-processors node. (In Russian).
Key words: HPC, supercomputer computations, fast Fourier
transform, FFT, GPU computations. |
|
article citation |
http://psta.psiras.ru/read/psta2017_1_47-62.pdf |
|
DOI |
https://doi.org/10.25209/2079-3316-2017-8-1-47-62 |
|
Article # 4_2017
20
p.
PDF |
submitted on 24th
Dec 2016 displayed on
website on 23th Jan
2017 Eugene Barkovksy, Ruslan
Kuchumov, Andrew Sokolov
Optimal control of two deques in shared memory with various
work-stealing strategies
In parallel load balancers based on work-stealing
strategy each processor has its own task deque. One end of the deque
is used by the owner to add and retrieve tasks, and the second — by
the other processors to steal tasks.
The aim of this research is to construct and analyze mathematical
models of the process of work with two cyclic deques located in the
shared memory. The parameters of these models are the probabilities
of operations (serial or parallel) at each step of discrete time.
Mathematical models are built as random walks on an integer lattice
in the plane. On the basis of models, problems of optimal partition
of memory were solved for various work-stealing strategies. As the
criterion of optimality we consider the maximum mean time to the
memory overflow.
Statistical studies to assess the probabilities of operations were
carried out for multiple types of tasks. For this purpose, as a part
of our RFBR grant, work-stealing balancer was constructed. Obtained
probabilities were used in numerical experiments to analyze the
developed models.
To solve these problems apparatus of controlled random walks,
absorbing Markov chains and LAPACK system were used. Calculations
were made using cluster KRC RAS. (In Russian).
Key words: work-stealing schedulers, work-stealing deques,
data structures, Markov chains, random walks. |
|
article citation |
http://psta.psiras.ru/read/psta2017_1_83-103.pdf |
|
DOI |
https://doi.org/10.25209/2079-3316-2017-8-1-83-103 |
|
Article # 5_2017
15
p.
PDF |
submitted on 28th
Dec 2016 displayed on
website on 23th March
2017 Vladimir Voloshinov, Sergey
Smirnov
Evaluation of a Coarse-Grained Branch-and-Bound Algorithm in the
Everest Computing Environment
The study concerned the
coarse-grained approach to the implementation of the parallel
branch-and-bound (B&B) method. Feasible domain of the mixed integer
programming problem is divided into several subsets
by xing some of the integer variables. It corresponds to
decomposition of the initial problem into subproblems,
which are being solved in parallel by a pool of
solvers. When solver nds an incumbent value it is broadcasted to
other solvers. Such exchange of optimal objective value’s
bounds allows to speed up the solution of
subproblems via reducing number of nodes of B&B search tree, which
are to be explored. Solver life cycle and incumbent values’
exchange are managed by the Everest Web-based
platform for distributed computing, everest.distcomp.org. The system
has been tested on several randomly generated
mixed-integer programming problems and a noticeable speedup has been
detected. (in Russian).
Key words: discrete optimization, branch and bound algorithm,
coarse-grained parallelism. |
|
article citation |
http://psta.psiras.ru/read/psta2017_1_105-119.pdf |
|
DOI |
https://doi.org/10.25209/2079-3316-2017-8-1-105-119 |
|
Article # 7_2017
15 с.
PDF |
submitted on 22th
Dec 2016 displayed on
website on 13th
March
2017
Aleksei Kondratev, Artem Bezzubtsev, Aleksandr Smirnov
The use of a distributed data processing system in the task of
building an automated surveillance system
Modern approaches to building surveillance systems
are beginning to increasingly rely on automation. This article
describes an approach to the construction of an automated video
surveillance systems based on sensors constructed using a
microcomputer, and a high-performance cluster calculator. At the
beginning of the work describes the basic approach to the collection
and processing of information. The main part describes the speci$cs
of the distributed data processing system, as well as the methods
and data collection and analysis algorithms. The $nal part describes
the results of experimental studies, as well as the major $ndings,
both positive and negative aspects of the approach and developed a
complex. (In Russian).
Key words: distributed systems, high-performance cluster
calculator, sensory compute node, Raspberry Pi. |
|
article citation |
http://psta.psiras.ru/read/psta2017_1_135-149.pdf |
|
DOI |
https://doi.org/10.25209/2079-3316-2017-8-1-135-149 |
|
Article # 9_2017
21
p.
PDF |
submitted on 18th
Dec 2016 displayed on
website on 17th March
2017 Anastasiya Belyaeva, Eugeniy
Biryaltsev, Marat Galimov, Denis Demidov, Aleksandr Elizarov, Ol’ga
Zhibrik
Architecture of HPC clusters for Oil&Gas Industry
We consider design and architecture of a complex integrated system
for high performance computing, data storage, and 3D visualization
targeted at oil and gas industry.
We describe technical details and results obtained while creating
such an integrated software and hardware solution, and talk about
challenges and directions for further progress in the technological
field. (In Russian).
Key words: numerical simulation, distributed computing, GPU
cluster, oil and gas industry, HPC platform. |
|
article citation |
http://psta.psiras.ru/read/psta2017_1_151-171.pdf |
|
DOI |
https://doi.org/10.25209/2079-3316-2017-8-1-151-171 |
|
Article # 10_2017
19
p.
PDF |
submitted on 20th
Dec 2016 displayed on
website on 17th March
2017 Sergey Abramov, Vladimir Roganov,
Valerii Osipov, German Matveev
MetaStochastic adaptive algorithms and theirs implementation with
parallel programming system T++ & MPI
The article describes parallel adaptive method for calculating
multidimensional integrals. (In Russian).
Key words: dynamic parallelization, T-system with an open
architecture, OpenTS, T++ programming language, adaptive algorithm,
Meta-Stochastic process, scalable computing, multiparameter system,
multidimensional integral, Monte Carlo method, recursive descent. |
|
article citation |
http://psta.psiras.ru/read/psta2017_1_173-191.pdf |
|
DOI |
https://doi.org/10.25209/2079-3316-2017-8-1-173-191 |
|
Article # 11_2017
16
p.
PDF |
submitted on 24th
Dec 2016 displayed on
website on 17th March
2017 Anton Baranov, Dmitriy
Lyakhovets
The influence of packaging on efficiency of parallel jobs
scheduling
The article considers the system of parallel jobs scheduling
designed by the authors. The system allows combining into one
package the same type of parallel jobs with the long initialization
time. Long-time initialization results into decrease of
efficiency of computing resources usage and parallel jobs scheduling.
The article presents the results of the experiments on the research
of the influence of packaging on such scheduling efficiency metrics
as full and useful load of computing resources. (In Russian).
Key words: jobs packaging system, batch mode, grouping-based
scheduling, SUPPZ, simulator, scheduling efficiency. |
|
article citation |
http://psta.psiras.ru/read/psta2017_1_193-208.pdf |
|
DOI |
https://doi.org/10.25209/2079-3316-2017-8-1-193-208 |
|
Article # 4_2017
20
p.
PDF |
submitted on 24th
Dec 2016 displayed on
website on 23th Jan
2017 Eugene Barkovksy, Ruslan
Kuchumov, Andrew Sokolov
Optimal control of two deques in shared memory with various
work-stealing strategies
In parallel load balancers based on work-stealing
strategy each processor has its own task deque. One end of the deque
is used by the owner to add and retrieve tasks, and the second — by
the other processors to steal tasks.
The aim of this research is to construct and analyze mathematical
models of the process of work with two cyclic deques located in the
shared memory. The parameters of these models are the probabilities
of operations (serial or parallel) at each step of discrete time.
Mathematical models are built as random walks on an integer lattice
in the plane. On the basis of models, problems of optimal partition
of memory were solved for various work-stealing strategies. As the
criterion of optimality we consider the maximum mean time to the
memory overflow.
Statistical studies to assess the probabilities of operations were
carried out for multiple types of tasks. For this purpose, as a part
of our RFBR grant, work-stealing balancer was constructed. Obtained
probabilities were used in numerical experiments to analyze the
developed models.
To solve these problems apparatus of controlled random walks,
absorbing Markov chains and LAPACK system were used. Calculations
were made using cluster KRC RAS. (In Russian).
Key words: work-stealing schedulers, work-stealing deques,
data structures, Markov chains, random walks. |
|
article citation |
http://psta.psiras.ru/read/psta2017_1_83-103.pdf |
|
DOI |
https://doi.org/10.25209/2079-3316-2017-8-1-83-103 |
• Содержание выпуска •
• Mathematical Foundations of Programming •
• Software and Hardware for Distributed Systems and Supercomputers •
• Artificial Intelligence, Intelligence Systems, Neural Networks •
• Supercomputing Software and Hardware •
• Software and Hardware for Distributed Systems and Supercomputers •
|
.en.html_cmp_strtedge110_vbtn.gif){kind=small}
/_derived/content-02(033).en.html_cmp_strtedge110_vbtn.gif){kind=small}
/_derived/content-03(034).en.html_cmp_strtedge110_vbtn.gif){kind=small}
/_derived/content-04(035).en.html_cmp_strtedge110_vbtn.gif){kind=small}
