2013 г. http://ea.donntu.ru/handle/123456789/29699 2026-04-22T22:48:32Z 2026-04-22T22:48:32Z Иванов, Ю.А. Іванов, Ю.О. Ivanov, Y.O. http://ea.donntu.ru/handle/123456789/29931 2016-03-09T14:13:22Z 2013-01-01T00:00:00Z

Title: Оптимизация вычислительного процесса на многоядерной системе с использованием алгоритма искусственной пчелиной колонии Authors: Иванов, Ю.А.; Іванов, Ю.О.; Ivanov, Y.O. Abstract: В статье рассмотрена задача разработки расписаний для эффективного управления процессом моделирования разночастотных динамических объектов в режиме реального времени на многоядерной платформе. Методами математического программирования данная задача классифицирована как смешанная нелинейная целочисленная. В работе было предложено решение с использованием метавэвристического алгоритма искусственной пчелиной колони, адаптированного для поставленной дискретной задачи оптимизации. Предложен подход для инициализации начального состояния роя, как решение задачи о камнях с использованием жадного алгоритма. Для рассмотренных наборов входных данных определен оптимальный состав популяции в алгоритме. Выполненная оценка работы и надежности алгоритма показала возможность применимости и высокий уровень его эффективности при решении задачи разработки расписания. Description: In this paper job-shop model for multicore platform was considered. The aim of the work is determination of model parameters and use them to develop schedule for hard real-time simulation. Relations in the model are set to achieve optimal control for dynamic systems including significantly different object frequencies. The model that describes computational process is complex and has several strict restrictions. In terms of mathematical programming the task is classified as a mixed integer nonlinear problem (MINLP). Given problem is multi-objective and has multiple extremas. That means finding a relative minimum is very complex for this task. Well known methods were considered: simplex method, simulated annealing, genetic algorithm, ant colony optimization and the artificial bee colony. Metaheuristic artificial bee colony algorithm was proposed for solving. This algorithm has been adapted for the solution of discrete optimization problem. In the paper a new approach was proposed to initialize the initial state of the swarm. Its value was determined as a solution of the bin packing problem using a greedy algorithm. In the worst case random initialization of a particular task may not give a feasible solutions that would satisfy the constraints posed. In addition a further bees movement to random positions also may not bring success. Therefore, as an initial value were proposed uniform basic task flow distributions across multiple cores to initialize bees’ positions in the work. With this formulation, we achieved the goal of the minimum value of the load on each the core. Also this solution allows getting an optimal value of the objective function on the first step of the algorithm. The optimal bee population for algorithm was defined for the known data sets. Experiment showed that algorithm is less sensitive to increase number of threads in simulation problem. In a case when required a schedule for specialized computer systems with more cores than the standard should be adjusted in the parameters of the algorithm to increase the performance and its accuracy. The reliability and feasibility of the algorithm were estimated for input test sequences. Test runs of algorithm demo nstrated its applicability and high efficiency schedules as a result.

2013-01-01T00:00:00Z Грищук, Т.В. Романюк, О.Н. Яремчук, В.В. Романюк, А.Н. Gryshchuk, T.V. Romaniuk, A.N. Yaremchuk, V.V. http://ea.donntu.ru/handle/123456789/29930 2016-03-09T14:10:58Z 2013-01-01T00:00:00Z

Title: Модель агента інтерфейсу, що реагує на зміну параметрів оператора Authors: Грищук, Т.В.; Романюк, О.Н.; Яремчук, В.В.; Романюк, А.Н.; Gryshchuk, T.V.; Romaniuk, A.N.; Yaremchuk, V.V. Abstract: У статті викладено основні принципи побудови комп’ютерної системи, що реагує на зміну параметрів оператора. Запропоновано математичну модель агента інтерфейсу технічної системи та розроблено програмне забезпечення. Description: The article describes the basic principles of a computer system that responds to changes of the operator's parameters. The article includes a short review of avatar’s models. These models have the specialized application and need simplification and generalization to be applied to technical control systems. The main drawback of the existing approaches is the lack of feedback interaction between the interface agent and the operator. The purpose of the research presented in this article is to increase the human-machine systems efficiency at the stage of data perception from operator. To achieve this goal, the authors propose to add a special module - an agent, that will carry on a direct "dialogue" with the operator. Features of the agent: tracking operator movements, analysis of operator’s voice data, providing instructions to operator concerning his/her activities in natural language. The system structure consists of the interface part, text-to-speech module, dialogue logic module, language recognition module and hardware. The hardware consists of an agent monitor, microphone, speakers and web-camera. The agent should also change its actual position on the monitor to increase the intuitive component of the human-machine communication. The mathematical model of the technical system interface agent is provided. The following operator parameters should be measured for this model: distance between the operator and the agent, horizontal and vertical shifts of operators from the agent, speech loudness. The agent mathematical model includes the agent’s shape scale, horizontal and vertical rotation angles, loudness of the agent’s voice instructions. The agent’s parameters depend on the operator’s state. The authors developed C++ computer program called Personal Assistant. For all the agent’s modules appropriate classes were developed. The tests were divided into three groups: GUI, verification of the system settings, and proofing the agent’s mathematical model adequacy. The following programming modules were used: SDK faceAPI, which allows tracking the movement of the human head, TTS module "Govorilka", which si mplifies the usage of the Microsoft Speech API, and NAudio library, which greatly simplifies access to audio devices and incorporates lots of useful built-in functions for sound analysis. The model of technology system agent, as opposed to others, takes into account the parameters of the operator and improves the efficiency of the human-machine systems on the stage of data perception from the operator.

2013-01-01T00:00:00Z Дмитриева, О.А. Дмитрієва, О.А. Dmitrieva, О.А. http://ea.donntu.ru/handle/123456789/29929 2016-03-09T14:10:04Z 2013-01-01T00:00:00Z

Title: Параллельный контроль размера шага вложенными методами на основе преобразованных матриц коэффициентов Authors: Дмитриева, О.А.; Дмитрієва, О.А.; Dmitrieva, О.А. Abstract: Статья посвящена вопросам параллельного управления шагом интегрирования при моделировании динамических объектов, описываемых системами обыкновенных дифференциальных уравнений (СОДУ) большой размерности. Для возможного управления шагом интегрирования предлагается использование параллельных вложенных методов с генерацией коэффициентов заданного порядка точности. Расчетные схемы методов построены путем приведения матриц расчетных коэффициентов к диагональному или треугольному виду, что позволило значительно сократить число обменов на шаге. Разработаны параллельные алгоритмы вариации шага с фиксированным числом итераций и основанные на контроле локальной точности. На основе построенных алгоритмов выполнена параллельная реализация тестовых задач. Description: The article deals with the parallel control integration step in the modeling of dynamic objects, described by systems of ordinary differential equations (SODE) of large dimension. Step control for the possible integration of the proposed use of parallel methods for the generation of embedded coefficients a given order of accuracy. Payment schemes are constructed on the methods of reducing a matrix of the estimated coefficients to the diagonal or triangular form, thus significantly reducing the number of exchanges in step. Developed parallel algorithms variations step with fixed number of iterations based on the control local accuracy. On the basis of these algorithms made parallel implementation of the test problems. Applying such techniques for parallel simulation of dynamic objects than high levels of parallelism is also characterized by good stability properties and high-order methods. Obtained on the basis of this approach, numerical schemes have less computational complexity due to the significant reduction in the number of exchanges and are very effective in solving stiff equations. The proposed approaches are based on modifications of the implicit multistep methods, concurrent receipt values in the stage-points, while sharing the values of the processors do not perform after each iteration, and then get the value for the next target point. Such a drastic reduction in the number of exchanges (s times, wheres - the number steps of the method) is achieved by using a diagonal or triangular approximation of the original matrix. For classical implicit techniques which are used to solve the equations and rigid systems as defined diagonal D or triangular T occurrences performed once, before calculation. For the modification of methods used in the software system Mathematica Wolfram Research, Inc.

2013-01-01T00:00:00Z Чередникова, О.Ю. Иванов, А.Ю. Коцогуб, Е.В. Іванов, О.Ю. Коцогуб, Є.В. Cherednikova, O.Yu. Ivanov, А.Yu. Kotsogub, E.V. http://ea.donntu.ru/handle/123456789/29928 2016-03-09T14:07:27Z 2013-01-01T00:00:00Z

Title: Исследование многопоточности с использованием волокон на одно- и многоядерном процессорах Authors: Чередникова, О.Ю.; Иванов, А.Ю.; Коцогуб, Е.В.; Іванов, О.Ю.; Коцогуб, Є.В.; Cherednikova, O.Yu.; Ivanov, А.Yu.; Kotsogub, E.V. Abstract: В данной работе рассмотрены принципы организации многопоточности с использованием волокон на одно- и многоядерном процессоре. Сравнивается организация многопоточности на основе потоков и волокон. Исследуется обращение из различных потоков и волокон к общей памяти. Предпринята попытка определения процессорного времени, выделяемого каждому потоку операционной системой. Аргументируются преимущества использования волокон для ускорения работы приложения на многоядерном процессоре. Description: In this work, the principles of organization of multithreading using threads and fiber to single- and multi-core processor are considered, their comparative characteristics are carried out. Using multithreading can greatly speed up the work of application, especially on multi-core processor. On single-core processor it is not possible to realize the true parallel work of threads. Each thread allocates some CPU time, after which there is a thread switching. It is established that the Windows 7 operating system allocates 94-96 msec for each thread. Moreover, if there are more than two threads with the same priority, it is impossible to reliably predict which of the threads to be executed. If the threads have different priorities, then run a higher priority thread. In this regard, of particular relevance gets fibers, because switching fibers is initiated by the programmer in the desired points in time, in contrast to switch threads that are switched by the operating system. However, on one thread can't run in parallel several fibers. Fibers can run in parallel only on different threads. In the work we studied the possibility of calling from various functions of fiber and thread to the shared memory. As have shown results of program research on the processor Intel Core 2 Duo, both threads can simultaneously access a single memory cell almost without delay, which can be useful when implementing parallel fragments of the program. This is because in today's multi-core processors L2 cache is a memory of separate use and shared on two cores, thus considerably simplifies the exchange of information between the cores, reducing the delays in the work of both cores with the same set of data. Further research is related to the use of fiber for the organization of protection against unauthorized access to information. Such use of fibers based on the fact that if fiber function will perform the identification of the computer running in place, to break this protection would be rather difficult, because the attacker is quite difficult to estimate the moment when this fiber function will be called.

2013-01-01T00:00:00Z