Software and Hardware for Distributed Systems and Supercomputers

Article # 31_2015

26 p.

PDF

Dmitriy Zmejev, Nikolay Levchenko, Anatoliy Okunev, Aleksandr Stempkovsky
The principles of organization of the I/O system of the parallel dataflow computing system

This article discusses the basic principles of I/O block of the parallel dataflow computing system “Buran”, analyses the effect of different algorithms for data entry on the efficiency of task computation. Describes a means to reduce the required size of content-addressable memory of keys and analyzes the results of experiments on the programming model of the parallel dataflow computing system. (In Russian).

Key words:  I/O block, algorithms of data entry, content-addressable memory, parallel dataflow computing system.

article citation

DOI

https://doi.org/10.25209/2079-3316-2015-6-4-3-28

Article # 35_2015

18 p.

PDF

Vladimir Roganov, Anton Kuznetsov, German Matveyev, Valeriy Osipov
Adaptive analysis of passwords’ reliability using computational power of hybrid supercomputers

The article concerns the security of traditional ways of password storing and handling. It describes known incidents of compromising password hash databases using high-performance computing clusters. It is noted that very often the cause for database vulnerabilities is an obsolete way of storing and handling passwords by users and system administrators. The research shows that adaptive cryptanalysis algorithms introduce the considerable security threat for password-storing information systems. Authors give security recommendations so as to help prevent intelligent cryptanalysis performed on HPC hybrid cluster systems. (In Russian).

Key words:  T-system, dynamic computing, T++programming language, cryptanalysis, information security, Markov chains, adaptive password searching, HPC.

article citation

DOI

https://doi.org/10.25209/2079-3316-2015-6-4-139-155

Article # 37_2015

19 p.

PDF

Alexei Adamovich
Fibers as the basis for the implementation of the notion of the T-process for the JVM platform

The spread and the accessibility of modern parallel computing platforms shows the lack of an appropriate support level of the needs of software developers for the parallel applications implementation. At PSI RAS the approach is under development to parallelization of programs based on the use of the computation model named “self-transformation of computational network”. This paper discusses the various options for approaches to the implementation of the notion “T-process” –– one of the basic notions of the computation model –– for the JVM platform. The potential problems are analyzed associated with the implementation of the “ T-process” notion on the basis of the classical OS/JDK threads, and in the event of a lightweight threads are supported directly in the virtual machine code. The approach to the implementation of the T-processes is proposed based on the use of the concept of the fibers, i.e. lightweight threads, which are implemented outside the JVM code. The results of experimental comparison of different approaches to the implementation of the notion of “T-process” based on the use of fibers and classical streams are provided. We also analyze the effect of the use of fibers in the implementation of the “self-transformation of the computational network” computation model which is used in the ajl parallel programming language being developed for the JVM platform.(in Russian).

Key words:  programming languages implementation, parallel computations, JVM platform, automatic dynamic parallelization, threads, fibers.

article citation

DOI

https://doi.org/10.25209/2079-3316-2015-6-4-177-195

Article # 41_2015

69 p.

PDF

Genri Norman, Nikita Orekhov, Vasily Pisarev, Grigory Smirnov, Sergey Starikov, Vladimir Stegailov, Alexey Yanilkin
What for and which exaflops supercomputers are necessary in natural sciences

An approach is developed which permits to expose tasks which need supercomputers to be solved. Near future pre-exaflops supercomputers are included. The approach potential is shown for examples of topical tasks in mechanics, physics, chemistry and biology. Both atomistic modeling & simulation and mechanics of continua are treated. Torus topology advantages are demonstrated. Recommendations are suggested for procedures of design selection of new supercomputers which are planned. (In Russian.

Key words:  atomistic modeling, electronic structure, molecular dynamics, multiscale modelling, radiation aging, laser ablation, nucleation, polymers, electrochemistry, parallel efficiency.

article citation

DOI

https://doi.org/10.25209/2079-3316-2015-6-4-243-311