SIMULATION MODEL OF GRAPHIC MULTIPROCESSOR BASED ON THE PETRI NETS THEORY

This work addresses the problem of creating the simulation models of a graphical multiprocessor for carrying out computational experiments to determine the efficiency of using the parallel computing based on GPGPU (General-purpose computing for graphics processing units, non-specialized computing on graphics processors) in tasks of structural-parametric synthesis of big discrete systems based on evolutionary procedures. We propose to use the Petri netstheory as a mathematical tool.It has the parallelism property and allows describing discrete processes occurring both in the genetic algorithm and in the computing system itself.The development of a simulation model is carried out on the basis of the graphic multiprocessor module memory architecture taking into account the specifics of its work related to the ability to read, write and transmit data.In addition, we describe the feature of the arithmetic logic devices work, which are able to simultaneously execute one command over a set of data.When building the model we take into account a feature of graphic multiprocessors which allows to get a greater effect from usingthe parallel computing avoiding the branching and control blocks operation that slow down the multiprocessor (since their number is less than the calculators number), thereby forming the “narrow” places.The proposed simulation model of a multiprocessor unit based on the selected tool is implemented using the specialized software for simulation based on the Petri nets theory – PIPE 5.This software is distributed free of charge and has a wide range of instrumental and analytical tools, which greatly simplifies both the modeling process and the process of analyzing the obtainedmodels.The resulting model will provide an opportunity to evaluate the efficiency of using parallel computing based on GPGPU technology in solving the task of improving the performance of intelligent information decision support systems based on genetic algorithms adapted to the subject area.

1. Al-Mouhamed M., Khan A.H. Exploration of automatic optimisation for CUDA programming.International Journal of Parallel, Emergent and Distributed Sys-tems,2015, vol. 30,is. 4,pp. 309 – 324.

2. Petrosov D.A. Adaptacija geneticheskogo algoritma pri modelirovanii vychislitel'noj tehniki s izmenjajushhejsja strukturoj i naborom komponentov na osnove setej Petri.Voprosy sovremennoj nauki i praktiki. Universitet im. V.I. Vernadskogo, 2009,no. 6 (20),pp. 151 – 160.

3. Petrosov D.A. Primenenie parallel'nyh vychislenij v intellektual'nyh sistemah upravlenija.Informacionno-analiticheskie sistemy i tehnologii. Materialy V mezhdunarodnoj konferencii, 2018,pp. 24 – 29.

4. Hart W. E., Baden S., Belew R. K., Kohn S. Analysis of the Numerical Effects of Parallelism on a Parallel Genetic Algo-rithm. In IEEE (ed.): CD-ROM IPPS97. 1997, 8p.

5. Al-Dabass D., Vindlacheruvu P., Ev-ans D.J. Parallelism in neural nets.Parallel Algorithms and Applications,1997, is. 3-4,pp. 169 – 185.

6. Lomazova I. A. Resource Equivalences in Petri Nets, in: Application and Theory of Petri Nets and Concurrency. 38th International Conference, PETRI NETS 2017, Zaragoza, Spain, June 25–30, 2017, Proceedings. Ed. By W. van der Aalst E. Best. Vol. 10258: Lecture Notes in Computer Science. Switzerland: Springer, 2017,pp. 19 – 34.

7. Magergut V.Z., Ignatenko V.A., Bazhanov A.G., Shaptala V.G. Podhody k postroeniju diskretnyh modelej nepreryvnyh tehnologicheskih processov dlja sinteza upravljajushhih avtomatov.Vestnik BGTU im. V.G. Shuhova, 2013,no. 2,pp. 100 – 102.

8. Magergut V.Z., Rubanov V.G., Chuev A.S. Formalizacija i analiz diskretnyh organizacionno-tehnologicheskih sistem so strukturirovannymi agentami na in-dikatornyh setjah. Belgorod, 2016, 149 p.

9. Basavin D.A., Petrosov D.A., Ig-natenko V.A. Primenenie tehnologii GPGPU v zadachah sozdanija intellektual'nyh sistem podderzhki prinjatija reshenij.Vysokie intellektual'nye tehnologii v nauke i obrazovanii. Materialy IV Mezhdunarodnoj nauchno-prakticheskoj konferencii.Saint-Petersburg,2017,pp. 63 – 65.

10. Amdahl Gene M. Validity of the Single Processor Approach to Achieving Large-Scale Computing Capabilities. AFIPS Conference Proceedings,pp. 483 – 485.