Kharitonov S.F., Evseeva A.V., Koltsova E.A.

Ivanovo State Power Engineering University

Ivanovo, Russia

MODELLING OF AIR DIFFUSION USING CUDA PARALLEL PROGRAMMING TECHNOLOGY

One of the most important steps in design and construction of buildings is calculation of air flow force impact on them. Despite all the advantages of mathematical modeling usage it requires huge computational and time resources for calculations and information processing. Usage of parallel programming technologies increases productivity and reduces computing time.

The article aims at describing the development of computer software system that will simulate airflow within a complex of buildings using CFD methods and CUDA technology.

The objectives of this project are as follows:

· Investigate the trends of modeling techniques of liquids and gases.

· Compose a mathematical model of airflow.

· Implement parallel algorithm using CUDA technology.

· Implement graphical user interface for visualization of simulation processes.

· Conduct simulation experiments.

· Analyze results of the experiments.

The Institute of Numerical Mathematics of the Russian Academy of Sciences has developed a solution in the considered area. They have designed an application that uses parallel algorithm to calculate turbulent airflow in the ground layer of the atmosphere using MPI technology.

In this paper airflow is modeled in two-dimensional system using “Pressure-Speed” equation system. The equation system contains Navier-Stokes equations and continuity equation. The system also contains boundary. This system assumes calculation of fields by the following algorithm:

1. At the initial time point initial values of velocity and pressure are defined for each field.

2. Pressure values are calculated for a new time point using velocity values.

3. New pressure values are used in the Navier-Stokes equations to calculate new velocity values for the new time point.

These actions are repeated iteratively.

A graphical user interface was implemented to visualize sequential and CUDA algorithms. The graphical interface was implemented as a Windows Forms application using C# programming language.

The user interface supports the following functionalities:

· airflow visualization;

· ability to configure the complex of buildings;

· preservation of modeled areas as templates.

Several experiments have been conducted to check that the modeling system shows correct results. Airflow in Ivanovo State Power Engineering University premises was modeled during one of the experiments. The results of the experiment are present in Figure 1.

Figure 1

Performance testing was also conducted as part of this work. Figure 2 shows the calculation time dependence on the size of calculation area for sequential and CUDA algorithms. The graph shows 1000 iterations execution time. The sequential program was launched on a dual-core AMD 2.2-GHz processor. CUDA program was launched on the GeForce GT 610 graphics card.

Figure 2

Thus, usage of CUDA technology significantly reduces the calculation time (on average 10 times). It allows calculating for large areas reducing the time costs by several times.

The results of the simulated airflow process are similar to expected in given conditions. The program developed using CUDA parallel programming technology in combination with graphical interface can be used for simulation of airflow effect in a single building or a group of buildings.

References

1. Loitsiansky, L.G. Fluid Mechanics. - Moscow: Drofa, 2013

2. Filatov E.Y., Jasinski F.N. Mathematical modeling of flows of liquids and gases: Proc. allowance. - Ivanovo: State Educational Institution of Higher Professional Education "Ivanovo State Power University", 2007. - 84 p.

3. Boreskov, A.V. Parallel computing on the GPU. The architecture and the CUDA programming model. - Moscow: Publishing House of Moscow University in 2012.