The Design And Implementation Of A Scalable Computing Acceleration Middleware System For Desktop PSE

Currently, professional desktop PSE (Problem Solve Environment, problem solving environment) improves computing performance by a variety of parallel solutions,such as Start* P, Parallel MATLAB Toolbox and MATLAB pMATLAB which represent for the computing acceleration solution, and the other parallel PSEs (Problem Solving Environment) including NetSolve. Webflow and the others. These solutions effectively improve the performance of the desktop PSEs.With the emergency of the multi-core general-purpose processor, graphics processors which is designed under many-cores architecture, and a set of floating-point accelerators, and the development of super-computer technology, there are more challenges for parallel programming.To solve the difference between the different parallel hardware, we represent a new solution for the parallel programming, which is named high-performance computing for personal computing middleware system. The system to be divided into three-tiers, which is composed by the application layer, middle layer and the calculation of layer three. Application layer which provides the application with the PSE interface, wrap the computing tasks and data into some packages, and send the packages to the middle layer, and then get the result from the middle layer after the completion of the data; the middle layer redirection the computing task,data and results, and manage the back-end computing resources; computing task layer executes the computing according to the computing command by a high-performance computing power of the local hardware or remote super computer, and then return the result. The middleware layer separates the PSE and HPC hardware. The PSEs do not have to know whether computing is completed in the local or in the remote server, or weather by the CPU or by the GPU or by the floating-point accelerator.Under the direction of above system design, we implemented the system, which using MATLAB as front-end, and using GPU and MPI as computing acceleration back-ends. To make sure of the security of user's data, we encrypt the user information with RSA, and encrypt the computing command, data and result with DES, when we use MPI cluster as back-ends. Finally, we test the performance of the local acceleration and remote acceleration. And the result supports the effective of the system.