RUS | UKR | ENG | DonNTU> Master's portal DonNTU
Materials on the theme of master's work:

Biography

 

Abstract

 

Library

 

Links

 

Report about the search

 

Individual task

 

 

The author's abstract

On a theme of master's work:

" Construction MIMD-solvers of the equations on base "

MPI and OpenMP standards"

(The head: the senior lecturer of chair of COMPUTER Malcheva of the R.V.

The adviser: the assistant to chair of the COMPUTER Smagin A.N.)


Introduction. A substantiation of an urgency of a theme


Modelling of dynamic systems with the concentrated and distributed   parametres is an actual problem for all areas of technics and technologies. Last years methods and simulars are intensively used in real technical and technological projects for check of reliability of design decisions and reduction of terms of introduction of projects and development of operated dynamic systems by attendants act as that as the major factor which guarantees quality of designing. Playing so important role in technical and technological progress, modelling as the research method puts forward a number of requirements to computing systems and their software as to implementers of models. The analysis shows, that actual remain working out focused to program parallel modelling (»«»¼ß environments in which important the role plays MIMD-reshatel the equations. The important problem is working out parallel desiders of the equations for dynamic network objects with the concentrated and distributed parametres. Problem orientation is in friendly for the user the description of object of modelling and solved problems, in a specific kind of results of modelling, and also in the general drawing up of the interface of the user.

The purpose and work problems

The purpose of the given work is working out, debugging and experimental researches MIMD-solver of the equations on the basis of MPI andMPI и OpenMP the standards, focused on the decision of problems of modelling of mine ventilating networks (MVN).

For achievement of this purpose in work such problems are solved:

The characteristic of network dynamic systems as objects of modelling.

Working out of algorithms MiMd-solver of the equations on the basis of MPI and OpenMP standards for model NDO(network dynamic object) with the concentrated parametres.

Implementation and experimental researches MIMD - desider of the equations on the basis of MPI and OpenMP standards for NDO with the concentrated parametres.

Prospective scientific novelty


To carry out the analysis of different approaches to construction MIMD-solver of the equations for modelling of network dynamic systems with the concentrated parametres that opens positive prospects of an effective utilisation of parallel computing resources for the complex decision of problems of modelling MVN.

Experimentally to confirm working capacity of algorithms the decision of the equations for networks of various complexity.

To create parallel решатели the equations which are components the parallel modelling environment that is a certain payment in development of parallel modelling environments for modelling of network dynamic objects.

Planned practical value

Practical value of work consists in working out:

Efficient and effective components of the parallel modelling environment, MIMD-solver the equations for NDO with the concentrated parameters which allow to solve practical problems of modeling MVN (design, operational, problems of automation of management);

The software package in language C++ with use MPI and OpenMP, which is a part of software-hardware of the parallel modelling environment which can be applied in the coal industry and   in educational process.


That is made to

Bases of modelling support of projects DS(dynamic system) have been appreciably fulfilled on analogue and analogue-digital simulars. Intensive transition has begun with the middle 80 to means of digital modelling: are developed and implementation on all types of digital COMPUTERS blochno - and the equation-focused languages of modeling DS, the hardware base of modelling - the personal and super COMPUTER, parallel computing systems SIMD - and MIMD-architecture which have been urged to compensate absence of super-power parallel processors in modeling complexes quickly developed.

The concept of distributed parallel modeling DS environment and the structural organisation of the parallel modelling environment which basic components are hardware, the system and modelling software have been in the mid-nineties developed. To hardware of environment access probably to have as from the workplaces territorially approached to super-power computing systems (superSISD, SIMD, MIMD), and with essentially kept away. In the latter case it should be used Internet technology of access. The structure provides local servers which can have parallel resources, the focused peripheral equipment and construction tools of seminatural modelling complexes is modelling.

Active participation in working out SIMD-components DPSE( distributed parallel simulation environment) the Stuttgart university took part. The structure of the modelling software has been in common developed, parallel algorithms of modelling DSCP(dynamic system with concentrated parameters), DSDP ( dynamic system with distributed parameters )are investigated, the dialogue subsystem is developed, a number of test problems is solved. Implementation algorithms it is executed in SIMD-language of parallel programming Parallaxis, and experimental researches - on SIMD-system MasPar. Works were conducted in such directions: generalisation of experience of parallel modelling at level of programming languages and working out of SIMD-language of modelling in interrelation with construction of a subsystem of dialogue; research of aspects of the object-oriented approach; construction of libraries of parallel modelling ДСКП, ДСРП; experimental researches problem the focused modelling SIMD-environments; application of parallel SIMD-models in mining and educational process. Last years working out MIMD-components is actual.

The problem of construction РПМС is a subject of scientific cooperation of faculty VTI with institute of the parallel and distributed super-power computing systems (IPVR) the Stuttgart university. Working out the focused to problem environments is carried out with institute of system dynamics and automatic control (ISR) the Stuttgart university and "Nitrogen" (Severodonetsk). The problem of integration of modelling environments with CAD is investigated together with automation and software institute (IAS) the Stuttgart university.

Current and planned results


It is at present developed:

Approaches to распараллеливанию algorithms of functioning ES(equation solver):

-         Based on MIMD to structure,   

-         Based on SPMD   to structure.

  Search of variant MIMD - распараллеливания can be conducted within the limits of these   program structures.

Process Т11 starts performance of processes. To the beginning of possible parallel performance Т1 "leading" process carries out the big number of consecutive operations and prepares for start operating streams Т 2 … Тn.

Further these processes work in parallel and reach everyone the termination of the program and operation of association which conducts to synchronisation with process Т1 and to terming of streams Т2 … Тn is carried out. At each performance of a parallel code in the program operating streams are generated and named again   (fig.1.11).


                                                                      Fig. 1.1. The MIMD-parallel approach

Generating and determine  operating streams causes expenses of resources. Their economy is reached in SPMD - models parallel.
In SPMD - models parallel processes (operating streams) are generated only once at program start. These processes are named in the end of program performance (fig. 1.2).
Consecutive parts of the program are carried out by all processes, i.e. it is a question of reusable performance of consecutive codes of the program over the local data.

 

Fig. 1.2 SPMD-approach to parallelising

 

          The analysis of these structures shows, that it is necessary to be guided by the SPMD-organisation solver to carry out the program on several MIMD-processors. Thus in each processor there will be initial and final data.

Analyzing variants  parallel  operations on calculation of components X, Y, it is possible to offer a number of variants virtual MIMD - models of network objects. As the virtual we will understand model in which in parallel functioning, rather independent processes in common providing the decision of a problem are allocated.
The scheme of virtual model which распараллелена on SPMD - to a principle also corresponds to quantity of branches, it is shown on fig. 1.3.

Fig. 1.3. Virtual MIMD-model ES (X and Y are not united)  

In this model it is provided m virtual processes, each of which conducts calculation or components X, or Y. After a finding of current values components X calculation of values of right parts Y components in which course there is an information interchange between processes is carried out, further elements Y - components are transferred to virtual processes components X.

Devirtualisation we name process of display of virtual model (fig. 1.4, fig. 1.5) on the target parallel computing system (TPCS), i.e. that computing system which is at the disposal of the user. We will execute  devirtualisation  for virtual model (fig. 1.3). As it is possible to see from fig. 4.5, Х - processes stand idle time part, besides, quantity of operations which is carried out by these processes, smaller, than at Y - processes. There is a sense to execute association Х - processes in groups that will make to X component, and Y - processes - Y to a component, on purpose to level loading of processors two component (fig. 1.4).

Fig. 1.4. DevirtualisationMIMD-models ES (X and Y are not united)

During the further work it is planned to realise the received approaches by means of OpenMP technologies and to analyse solver perfomance the equations.

The conclusion

Intensive development of means of information technologies for modelling of difficult dynamic systems gives possibilities for reduction of material inputs and human resources. During the given work it is developed MiMd-solver the equations on the basis of MPI and OpenMP standards. To reach the purposes put before the developer, it is necessary to develop an effective technique of realisation parallel solvers the equations with use of leaders MPI and OpenMP standards. .