Processors with multiple CPU cores have become widely available. Therefore it is useful to parallelize the Monte Carlo simulation process in the context of metrology. Different approaches for parallel computing including Monte Carlo simulations are available and have been studied. We choose an approach which is based on a client-server model and which allows the use of remote computing resources when they are available. The communication between clients and servers is socket based using a high level remote procedure call protocol which integrates different system architectures (e.g. Windows and Linux) in an overall simulation network. The user can access the simulation network via a lightweight client which delegates all calculations to his network entry server. The results are communicated back to the client when the simulation task is finished. The system is optimized to limit the communication between client and server to a minimum and it is buffered to allow the calculation processes to continue, even if communication might be slow. The concept is flexible enough to be used by computers with multiple CPU cores or by multiple computers connected by a fast local network (LAN) or any combination of both.
Citation: Advanced Mathematical and Computational Tools in Metrology and Testing (AMCTM-2011)
Pub Type: Journals
Monte Carlo Simulation, parallel processing