Physical modeling of real world phenomena often demands significant computer resources, such as speed and memory, that far exceed that offered by even the largest single processor or shared memory systems. The national laboratories including Lawrence Livermore, Sandia and Los Alamos, are using large distributed computer networks to handle these problems. Parallelization of an algorithm across a distributed network is not a mechanized process and the approach differs drastically between problems. Fortunately, some of these problems can be classed, and generic solutions applied to them. One class of these problems is radiation transport problems.
Radiation transport using the Monte Carlo method is well-known and is used extensively at Lawrence Livermore National Laboratory (LLNL). To accurately model a sophisticated three-dimensional problem including detailed geometry specification it is common to represent the problem with a mesh. The resolution can be controlled by the number of zones in the mesh which dictates the size of the problem, that is the amount of memory needed to store the mesh. As the mesh sizes become large it will not fit on even the largest single memory unit. (Note that even if it could, we would still want to run an even larger problem simply because we can - i.e. distributed memory...) Therefore, in order for the Monte Carlo particle packages to handle problem descriptions larger than conventional memory on a single shared memory node, the problem must be distributed across multiple nodes. A distributed network computer must then domain decompose the mesh such that it can be read into memory. Using a distributed network computer requires the application to communicate between nodes using message passing techniques such as with the MPI library. Learning the details of distributed computing requires time and experience not common to many developers of particle transport packages. In order to provide developers with an intuitive interface to efficiently handle domain decomposition and particle passing, we propose development of the Parallel Particle Framework.