In modern conditions the pace of technological progress and successes in finding solutions of the fundamental scientific issues are strongly determined by the level of computing technology. At this time, a class of fundamental scientific and engineering problems Grand challenges is generally known, an effective solution of which is possible only with the use of powerful computational resources with a perfomance of hundred Gflops (~ 1012 operations per second) and above.
To justify the need for the performance, for example, for aerogas dynamics tasks, it is possible to provide relatively simple calculations of aerodynamic characteristics of an isolated wing. It can be shown that for stationary solutions of Navier-Stokes equations it is necessary about 2∙1013 arithmetic operations. If you are using a modern personal computers, it will take tens of hours. In practice, the aerodynamic design as usual does not need a separate calculation, but a systematic enumeration and analysis of many alternatives. Hence it is clear that only with the computer perfomance, which significantly exceeds the performance of personal computers, numerical modeling can be a real tool for the design of aircraft. The only way to achieve the necessary performance at this time is the use of multiprocessor computing systems.
However, in the case of multiprocessor computer systems it is much more complex to achieve high performance than on traditional computers, in dealing with real applications. The main requirement to the algorithm is the existence of internal parallelism. This means that the algorithm should be compiled from a number of parts that can be run simultaneously and independently of each other. The next principle determines the possibility of effective implementation of parallel algorithms. For multiprocessor systems, duration of communications between processors is much higher than the access time to local memory and, moreover, the duration of the arithmetic operations. Hence the condition of locality of the algorithm appears - at each processing element (PE) access to a local memory and the execution of arithmetic operations must be significantly more frequent than exchanges with the second PE. In the end, it should be noted quite a desirable requirement of scalability, which means the ability of the algorithm to work on an arbitrary number of processors. In practice, this feature provides a highly efficient parallel implementation for a specific number of PE.
Today, thanks to a new high-perfomance technics, received a substantial advance in solving problems of computational mechanics, modern physics, quantum chemistry, biology and other fields of science.
Modeling CDS gets all a greater urgency. Multiprocessing systems are applied to realization of CDS-models, and all potential computing capacity of such system is used not completely. For achievement of greater efficiency of system developers of the parallel software solve a problem paralleling programs
In the time of parallel modeling of Network Dynamic Objects with the Distributed Parameters (NDODP) systems with a lot of the equations are solved. Thus there is a problem of distribution of computing loading between processes.
The concept of a distributed parallel simulation environment (DPSE) for complex dynamic systems with lumped and distributed parameters was proposed in 1992 in the framework of scientific cooperation of VTI DonNTU and institute of parallel and distributed systems (IPVS) Stuttgart University (Germany), published in Asim-report [1]. DPSE system is called such a systematic organisation of joint functioning of parallel hardware resources, system and the simulator software, which supports all phases of design, implementation and use of parallel models of SDS [2].
The main provision of DPSE-concept is the need to fully develop parallel methods and algorithms for operation of the simulator software (Modeling and Simulation Software) for DSLP, DSDP. The analysis shows that the parallel SIMD-systems and MIMD-structures of 90-ies have brand parallel programming languages, which are based on languages, Fortran, C, C++, Modula-2 and other. Development of parallel computing systems with MIMD-architecture, object-oriented approaches promoted the standardization of tools for parallel and distributed programming. Thus, ANSI and ISO have defined C++ standard with MPI, PVM and Pthreads libraries. The concept provides language and system-institutional means to users and developers of parallel modeling that are able to overpass the fifth level systems and modelling languages [3]. In this direction the developments of generalizations about the topologies of the CDS are undertaken, and the complex "Topological analyzers - generators equations-solver" translates the description of the CDS level subject area in parallel programs [2].
http://www.computerbase.de/lexikon/Parallele_Programmierung
Основные понятия и термины параллельного программирования.
Святний В.А. Паралельне моделювання складних динамічних систем // Моделирование - 2006: Международная конференция. Киев, 2006 г. - Киев, 2006. - С. 83-90.
Святний В.А., Молдованова О.В.,Чут А.М.: Стан та перспективи розробок паралельних моделюючих середовищ для складних динам_чних систем з розпод_леними та зосередженими параметрами. // Моделирование - 2008: Международная конференция. - Киев, 2008. - С. 25-37.
Schmidt B. Simulationssyteme der 5. Generation - SiP, Heft 1, 1994, S. 5-6.
Бройнль Т. Паралельне програмування: Початковий курс: Навч. посiбник / Переклад з нiм. В. А. Святного. - К.: Вища шк., 1997. - 358 с.
Хьюз К., Хьюз Т. Параллельное и распределенное программирование на С++ / Пер. с англ. – М.: Издательский дом «Вильямс», 2004. – 672 с.: ил.
Moldovanova O.V., Svjatnyj V.A., Feldmann L., Resch M., Küster U.: Problemorientierte parallele Simulationsumgebung. // Научные труды ДонНТУ, серия «Информатика, кибернетика и вычислительная техника», вып. 93. – Донецк, 2005. – С. 145–150.
Святний В.А. Паралельне моделювання складних динамічних систем // Моделирование – 2006: Международная конференция. Киев, 2006 г. – Киев, 2006. – С. 83–90.
Гусєва Г.Б., Молдованова О.В. MIMD-паралельний вирішувач рівнянь для мережного динамічного об’єкту з розподіленими параметрами // Проблемы моделирования и автоматизации проектирования динамических систем: Сб. научн. тр. ДонНТУ, вып.6, Донецк, 2007.
Бондарева Е. С. Генератор уравнений параллельной моделирующей среды для технологических процессов // Проблемы моделирования и автоматизации проектирования динамических систем: Сб. научн. тр. ДонНТУ, вып.6, Донецк, 2007.