Modeling a system with communicating concurrent processes is an effective way of exposing parallelism inherent in the application. When multiple processes are mapped to a processor, however, we have to serialize the execution of the processes. Such serialization usually requires scheduling that may introduce significant runtime overheads due to context switching. In this work, we propose techniques to synthesize software with low runtime overhead from the set of processes running on a processor. Particularly, we convert concurrent processes into a sequential C++ program with process scheduling implemented in the program and replace operating system''s communication API by user-level communication API. Since there is no code explosion in our approach, it is applicable to big applications and we show it through experiments with a real-world example.
展开▼
