As an anniversary computing technology for the simulation of rarefied gas flows, Direct Simulation Monte Carlo (DSMC) digital simulation method is widely applied in many high science and technology areas such as aeronautics and aerospace, micro-electronic mechanical system, space material and nanotech, etc.. DSMC puts up strict requirements on high speed and memory capacity for simulating computers, which single CPU computer systems can hardly fulfill.With the development of parallel computing, there emerged more and more kinds of high performance computing platforms. Carrying out Direct simulation Monte Carlo on these platforms is becoming a key research direction. It is of great theoretic and application importance to study parallel programming technology and tools for the DSMC.The traditional CFD parallelization technology based on continuous medium emphasizes the contiguousness of physical effects and takes fluid parameters at grid points as computing objects. Whereas, the parallel computing of DSMC based on discrete medium emphasizes the dynamic movement of molecules and treats parameters of moving molecules as computing objects. Distribution of the computing tasks of the former is much easier than that of the latter one, of which the computing tasks are changing with the movement of molecules in computing period. So the task distribution of CFD can be fulfilled by dividing the grids in a static way, but DSMC cannot. Task distribution can only be done in dynamic way.The research work in area of DSMC parallelization is still in the bud. At abroad, DSMC research is focused on parallelization of certain particular problems. It is far away to form a systemic theory and method. We do not find any information about the development of DSMC parallel programming tools. As for in China, there is almost no such literature published.Cooperating with Energy and Power Engineering School, Xi'an Jiaotong University, our laboratory has begun to explore and research the parallelization technology of DSMC programs for the first time. We have got elementary successful results.The data dependency of DSMC is caused by the random migration of molecules across the divisional line between sub-domains of different computing111nodes. This paper presents a new concept of migration-dependency, based on which the dynamic dependency analysis of the parallel sub-tasks can be accomplished by tracing the movement of molecules in the parallel simulation. This is the foundation of DSMC parallel computing model.A parallel computing model is presented, based on migration-dependency. A Scalable Parallel Run-time Library is developed to solve all the key difficult problems such as dynamic data dependency analysis, synchronization, communication, and load balancing.A mechanism of judging run-time communication based on the migration dependency is also presented. This mechanism solves the puzzle of the sustained change of communication quantity. A mirror-anti-direction communication method is brought forward to avoid communication deadlock and to program expediently.The paper puts up with a dynamic load balancing method based on data migration to resolve the load imbalance in the parallel simulation caused by the free movement of molecules. Moreover, a load information adaptive acquisition method in the information strategy and a heavy-load-first algorithm in the location strategy are presented. A good load balance result has been reached.The paper carried out the architecture of the DSMC interactive parallelization tool software. The architecture is integrated into the ready made one for CFD programs parallelizer -PARACTIVE. The result software tool is called as PARACTIVE+. Two DSMC programs from Xi'an Jiaotong University have already been parallelized successfully. Parallel efficiency obtained can achieve 85% when executed on 4 computing nodes, and 75% on 8 nodes cluster system, comprised of 9 Pentium IV 1.7G PCs. These results gained the positive evaluation from the user definitely.The paper gives a model for