A method or device may optimize applications on a parallel computing system. Environment variables data may be used as well as a test kernel of an application to optimize communication protocol performance according to a set of predefined tuning rules. The tuning rules may specify the output parameters to be optimized, and may include a ranking or hierarchy of such output parameters. Optimization may be achieved through use of a tuning unit, which may execute the test kernel on the parallel computing system, and may monitor the output parameters for a series of input parameters. The input parameters may be varied over a range of values and combinations. Input parameters corresponding to optimized output parameters may be stored for future use. This information may be used to adjust the application's communication protocol performance “on the fly” by changing the input parameters for a given usage scenario.