Parallel System Performance For The Cfd Program

The developing of the automatic parallelizer for CFD (Computational Fluid Dynamics) oriented programs has outstanding application future and important theoretic significance. Our team has accumulated much experience after several years of intensive research on this area and developed an interactive parallelizer.The main work of this paper is as follows:1. To assist the doctors to test the performance of PARACTIVE.2. To evaluate the performance according to the data and the parameters in parallelization.3. To analyze the factors that influence the performance and put forward the further research issues.We parallized practical programs in CFD using PARACTIVE. And we tested the parallel programs obtained on YH-III super computer, PC clusters and SGI workstation. The efficiency can achieve 80% when executed on 4 computing nods, and 70% on 8 nods of YH-III super computer. The data drawn from the testing of benchmarks show that PARACTIVF, has the equivalent performance as the PFA and AFT. Dr. Fu you developed PARACTIVF+, which has enhanced the functions of dynamic load balaneing, and parallel zed the programs supplied by. the School of Energy and Power Engineering, Xian Jiaotong University. I assorted her to debug the parallel programs obtained and the results are satisfactory.The performance test also showed that PARACTIVE still cannot fully meet the requirements for practical application, due to the lack of enough inter-procedural analyzing ability. Parallelization for certain user CFD programs is not effective. From the test data of top level parallel ization tools, we can see that their functions are also far from practical application. This means that the research on programparallelization is very challenging, with many obstacle ahead.PARACTIVE is still in trial stage, and further improvement is required. It is necessary to increase the system' s analyzing ability, expand the recognition scope, and collect more CFD programs in order to use different parallelization methods for different programming style and then extend the appicaltion scope of parallelization tools.