Українська   Русский
DonNTU   Masters' portal

Abstract on the theme of master's work

Content

Introduction

One of the main strategic directions of the modern computer industry is the use of high-performance parallel computer systems (CS). This trend is due to a fundamental limitation of the maximum possible performance of conventional sequential machines, as well as continuous development of new computational tasks for which the capabilities of existing computer equipment is always not enough. Simulation of real multidimensional dynamic processes described by systems of ordinary differential equations (SODE) is one of the classes of problems for which the use of parallel supercomputers is not only justified, but necessary. This is evidenced by a well-known list of big challenge in which such tasks occupy a leading position [1].

The maximum performance of existing parallel computing systems grows substantially. And at last years technological base and architecture has been radically changed. Despite this facts, today the main obstacle of almost all parallel architectures is the lack of efficient parallel algorithms and their implementations. Thus, the problem of improving the efficiency of parallel computing systems, despite some progress, is far from being fully resolved, both in theory and in practice.

The most difficult task is determining the optimal algorithmic and structural organization of parallel computing in order to achieve sufficient performance characteristics. At present, there is no doubt that to realize the potential of parallel computers is possible primarily on the basis of the theoretical work focused on the construction of new and the adaptation of existing methods for solving complex scientific and engineering problems. Therefore the prospects of development of supercomputers in the near future are determined by the success not so much electronics as computational mathematics [2].

Relevance of the topic

There is one important rule among the many laws of the computer world – the performance of computer systems improve steadily and continuously. The need to improve performance of computing is largely driven by practical necessity. In addition, the increase in performance computer technology allows to increase the complexity of the task and constantly expand the range of problems under investigation.

Productivity growth contributes to continuous improvement of technologies for the computers creation. Until recently, such an increase in productivity was provided largely by increase of the clock frequency of basic computing elements – processors. But the chances of such approach are not limitless – after some moment further increase of clock frequency requires considerable technological effort, accompanied by a significant increase in power consumption and insurmountable problems of thermoregulation. In such circumstances almost inevitable radical change in the basic principle of computer manufacturing was born – instead of creating a new complex of high-frequency processors focus shifted to the development of composite processor consisting of a plurality of equal and relatively simple computational elements – cores. Maximum processor performance in this case is equal to the amount of computing cores in processors. Thereby productivity growth without increasing of the clock frequency.

It can be argued that an era of multi-core processors has begun. Transition to multi-core computers simultaneously marks the advent of parallel computing. This implies that the parallel computing become inevitable and ubiquitous.

On the other hand there are the following problems: numerical methods in the case of multi-core system should be designed in terms of parallel and interacting processes, allowing the implementation on independent computational cores; applied algorithmic languages and system software should ensure the creation of parallel programs, synchronization and mutual exclusion in organization of asynchronous processes, etc. These problems of parallel computations increase the existing gap between the computing capabilities of modern computer systems and existing algorithmic and software. And as a result, the elimination and reduction of this gap is one of the most important problems of modern science and technology.

Thus, in spite of extensive research in the field of parallel computing, works in this direction do not lose their importance and require further development because of the massive proliferation of parallel computing systems and the lack of proper software component [3].

At the time of writing of this essay master's work has not completed yet. Final completion: December 2014. Full text of the work and materials on the topic can be obtained from author or his manager after that date.

References

  1. Grand Challenges: Science, Engineering and Societal Advances Requiring Networking and Information Technology Research and Development [Электронный ресурс] // Report by Interagency Working Group on Information the Technology Research and Development. — Режим доступа: http://www.nitrd.gov/pubs/200311_grand_challenges.pdf
  2. Забродин А.В. Параллельные вычислительные технологии. Состояние и перспективы // Материалы первой молодежной школы ”Высокопроизводительные вычисления и их приложения” — М.МГУ, 2000.
  3. Гергель В.П. Теория и практика параллельных вычислений // В.П. Гергель — М.: Мир, 2007 — 424 c.
  4. Хайрер Э., Нерсетт С., Ваннер. Г. Решение обыкновенных дифференциальных уравнений. Нежесткие задачи. // Э. Хайрер, С. Нерсетт, Г. Ваннер. — М.: Мир, 1990. — 512с. — ISBN 5–03–001179–X.
  5. Хайрер Э., Ваннер. Г. Решение обыкновенных дифференциальных уравнений. Жесткие задачи. // Э. Хайрер, Г. Ваннер. — М.: Мир, 1999. — 685с. — ISBN 5–03–003117.
  6. Голуб Д. Ван Лоун Ч. Матричные вычисления // Д.Голуб, Ван Лоун Ч. — М.: Мир, 1999. — 548 с.
  7. Холл Дж. Современные численные методы решения обыкновенных дифференциальных уравнений // Дж.Холл. — М.: Мир, 1979. — 312 с.
  8. Воеводин В.В., Воеводин Вл.В. Параллельные вычисления // В.В. Воеводин, Вл.В. Воеводин — БХВ–Петербург, 2004 — 160 с.
  9. Фельдман Л.П., Назарова И.А. Современные параллельные методы численного решения задачи Коши // Фельдман Л.П., Назарова И.А. — Донецк: ДонНТУ, 2013. — 206с.
  10. Чисельні методи в інформатиці: підручник для вузів / Лев Петрович Фельдман, Анатолій Іванович Петренко, Ольга Анатоліївна Дмитрієва; За заг. ред. М.З. Згуровський. — Київ: BHV, 2006. — 479 с.
  11. Михайлова Т.В. Оценка эффективности высокопроизводительных вычислительных систем с использованием аналитических методов // Материалы 2-й международной научно-технической конференции " Моделирование и компьютерная графика", г. Донецк, 10-12 октября 2007 г.
  12. Боресков А.В., Харламов А.А., Марковский Н.Д., Микушин Д.Н., Мортиков Е.В., Мыльцев А.А., Сахарных Н.А., Фролов В.А. Параллельные вычисления на GPU. Архитектура и программная модель CUDA: Учеб. пособие / А.В. Боресков и др. Предисл.: В.А. Садовничий. — М.: Издательство Московского университета, 2012. — 336 с.