Abstract

Content

• Introduction
• 1. Theme urgency
• 2. Goal and tasks of the research
• 3. The research of loopback test on SMP system
• Conclusion
• References

Introduction

The idea of parallel processing as a powerful reserve increased performance of computing devices was proposed by Charles Babbage in about one hundred years before the advent of the first electronic computer. However, the level of technology the mid-19th century did not allow him to realize this idea. With the advent of the first electronic computers, these ideas often became the starting point for the development of the most advanced and productive computing. No exaggeration to say that the whole history of the development of high-performance computing systems — a history of the ideas of parallel processing at some stage in the development of computer technology, of course, combined with an increase in the frequency of electronic circuits.

1. Theme urgency

Computer science in its development on ways to improve the performance of the computer closer to the physical limits. The switching time of electronic circuits has reached fractions of a nanosecond, and the speed of propagation of signals in the lines connecting the elements and components of the machine, limited to 30 cm/ns (speed of light). Therefore further reducing the time switch circuitry will not significantly improve the performance of computers. Under these conditions, practical requirements (complex physical and technical calculations, multidimensional economic and mathematical models, and other tasks) to further improve the performance of the computer can be met only by extending the principle of parallelism of information processing devices themselves and create multi-machine and multiprocessing computer systems. Such systems can parallelize the execution of a program or run several programs at the same time.

One variety of such systems is the SMP system. A characteristic feature of SMP multiprocessor architecture is that all processors have direct and equal access to anywhere in the shared memory. The first SMP system consisted of several homogeneous processors and shared memory array, which processor connects via a common system bus. But very soon it became clear that such an architecture is not suitable for creating large-scale systems. The first problem, which arise — a large number of conflicts when accessing a shared bus. The acuteness of the problem could relieve some division memory blocks with which communication is possible with the help of switches parallelize applications from different processors. However, this approach is unacceptable seemed large overhead for the systems over with 32 processors [1].

Master's work is devoted to the actual scientific problem — improving the efficiency and the expansion of the class of tasks through the development of special Scheduler software and monitor system performance for multi-core SMP system. Investigation of complex dynamic systems based on multi-core computer will analyze the existing conflicts with shared virtual memory, consider the characteristics of the organization of modern architecture cache memory, and develop parallel techniques of data flows and optimize performance in the complex dynamic system on multi-core SMP-platform.

2. Goal and tasks of the research

The aim of this study is to develop a special software scheduler and monitor implementation of the system for multi-core SMP computers.

The main tasks of the researching:

1. An analytical model describing the analysis of dynamic systems.
2. The development of the block diagram of parallel model of a dynamic system and improving the exchange of information in this model.
3. The implementation of a multi-frequency monitor control dynamical system.
4. The research of the temporal characteristics of computing and exchanges with standard programming tools.
5. Development of the scheduler and monitor multi-frequency dynamic control system.

The object of research: complex dynamic system for multi-core SMP-platform.

The subject of research: development of models and special scheduler software , expanding the class of tasks.

Scientific novelty consists in the synthesis and study of cyclic time schedule for dynamic systems.

As part of the master's work is planned to obtain relevant scientific results in the following areas:

1. Improving the efficiency of information exchange in the parallel model of a dynamic system.
2. The parallel technique of data flows with the small volumes.
3. Expanding the class of tasks on SMP-platform and method of creating programs for this class.

3. The research of loopback test on SMP system [5]

SMP — symmetric multiprocessing. There are two types of architectures of the systems of this kind. For our research was chosen multiprocessor architecture (Figure 1), in which two or more identical processors are connected to the shared memory. A characteristic feature of these systems is that all processors have direct and equal access to anywhere in the shared memory. The first problem, which arise — a large number of conflicts when accessing a shared bus. Therefore, there are alternatives to this system — are MPP system and architecture NUMA (where virtual memory is divided into separate blocks).

The harmonic oscillator — a system that, when displaced from its equilibrium position experiences a restoring force proportional to the displacement, according to Hooke's law[3] : F = -k*x, where k — the spring constant.

The equation of motion of the cargo with an x, according to Newton's second law can be written as:

To create a software model of GO performed numerical integration of the system of differential equations obtained by reduction of the order of the harmonic oscillator equation:

Multi-threaded implementation of the harmonic oscillator

There are such main groups of numerical methods for solving systems of differential equations: numerical methods of Runge-Kutt, numerical methods of Adams-Boshfort and numerical methods of Euler. We have chosen numerical method Euler formula trapezoids, because of the simplicity of program implemented in an iterative cycle. For us, there is not critical some error in this method.

To research the loopback test in a model used two harmonic oscillators with different natural frequencies. A model was developed, which included two systems of equations whose solution was used Euler method. Software implementation of the model means formed MSVS C++. At the same mechanisms used to work with threads WinAPI. Each of the oscillators implemented as a separate thread.

SMP implementation of the harmonic oscillator

In the research of SMP-systems [4], as a tool used by the system with quad-core Intel i7 3770-K. To implement the multi-core model used the API for streaming.

Conclusions

The researchs have shown performance multi-threaded programming model of the harmonic oscillator and can be used for temporary profiling custom applications and provide comparative performance computing systems.

Master's work is devoted to the actual scientific problem of the synthesis and study of cyclic time schedules for dynamic systems. As part of the research carried out:

1. Analysis of mathematical models describing the dynamical system.
2. The analytical solution of differential equations.
3. A series of experiments on the loopback test.

Further research is focused on the following aspects:

1. The implementation of a multi-frequency monitor control dynamical system.
2. The development of the scheduler and monitor multi-frequency dynamic control system
3. The research model and special scheduler software.
4. Improving the efficiency of information exchange in an SMP system dynamics model for the research of electric rolling stock.

In writing this essay master's work is not yet complete. Final completion: December 2013. Full text of the and materials on the topic can be obtained from the author or his manager after that date.

References

1. Букатов А.А., Дацюк В.Н., Жегуло А.И. "Программирование многопроцессорных вычислительных систем", Ростов-на-Дону, 2003 г.
2. Беки Дж.А. Теория и применения гибридных вычислительных систем / Беки Дж.А., Карплюс У.Дж // Перевод с англ. под ред. д.т.н., проф. Б.Я. Когана – М.: Мир, – 1970. – 483 с.
3. Материалы из свободной энциклопедии "Википедия" — Гармонический осциллятор [Электронный ресурс]. – Рeжим дoступa к ресурсу: http://ru.wikipedia.org/wiki/Гармонический_осциллятор.
4. Эхтер Ш. Многоядерное программирование / Ш. Эхтер, Дж. Робертс – СПб.: Питер, 2010. – 316 с.
5. Первусяк А.И., Иванов Ю.А. "Исследование кольцевого теста на SMP системе".
6. Магистерская работа Ю.А. Иванова "Исследование и разработка алгоритмов планирования процессов реального времени моделирования сложных динамических систем".
7. Ванг В., Мишр П. и Ранк С. "Динамическая реконфигурация кэша и разметка с целью оптимизации энергопотребления в многоядерных системах реального времени".
8. Санчез И., Гонсалез А. "Анализ локальности данных SPECfp95".
9. Иванов А.Ю. "Построение службы синхронизации процессов".
10. Информация о преподавателе Дмитриевой О.А.[электронный ресурс] / Сайт кафедры прикладной математики и информатики ДонНТУ, - http://pmi.donntu.ru/text/prepod/dmitrieva.html