The Design And Implementation Of Ridiation Heat Transfer Solver Under GPU/CPU Hybrid Architecture

The combustion chamber is an important part of aviation engines. With the rapid development of modern computer technology advances and scientific computing numerical simulation methods, computational fluid dynamics(CFD) in terms of numerical calculation engine combustion chamber is gradually playing an increasingly important role. In the past, the combustion chamber is designed with the experimental studies and empirical analysis method of combining, it takes such a long period of time, what?s more, the cost of the experiment is very high. Therefore, the using of CFD numerical simulation has become a fast, cost-effective method of engine design.In this paper, with the study of radiation heat transfer to the combustion chamber, based on the open source CFD software Open FOAM, a radiation heat transfer software was developed. The software is based on Monte Carlo ray tracing method, devised a set of numerical simulation method is suitable for the gas radiation heat transfer by changing radiation source term. And the simulation accuracy is improved.Firstly, based on the theory of ray tracing radiation heat transfer, a serial solver was developed, which implements the basic functions of radiation heat transfer simulation.Then, with the serial solver, analyze radiation ray transmission convergence rate of different conditions, and compare the characteristics of GPU / CPU hybrid platform architecture, using NVIDIA's CUDA programming framework, with the Using of nonblocking MPI protocol, develop a cluster of "fat computing nodes" and "thin computing nodes", the "fat" contains GPU. With the testing of the parallel solver, it burst up a 10 times computing speed.Finally, with DHCP/PXE and the NFS file system, set up a cluster which contains five fat computing nodes and two thin computing nodes. Because of all the files were stored in the main storage and all the other nodes boot from the network, the administrator has the ability to configure the cluster OS/application in an easy way. And there are no need to reconfigure a node OS and application when adding or deleting it.The solver was running rapidly on the cluster. Through actual verification, the radiation heat transfer solver shows a good portability, also can be used to solve specific problem, and the computing speed has meet the design requirement.