Thread Pool Parameters Tuning Using Simulation

Abstract

The problems of multithreaded programming are discussed. One way to improve the quality of writing a parallel program is to use a model which can represent the process of parallel computing in details. In this research construction of model of the multithreaded algorithm is carried out using Petri-object simulation technology grounded on stochastic Petri net and object-oriented approach. Synchronous and asynchronous performance can be reproduced in model according to the resources availability. The parallel computation is reproduced with taking into account not only formal rules of threads interaction but also the time delays of program instructions. Due to that simulation results are close to the real which has been obtained by launching a multithreaded program. In particular, the model of a thread pool as one of the most efficient high-level tools of parallel programming is developed. The experimental results show speedup dependence on algorithm complexity, computing resources and parameters of the thread pool. This dependence is revealed both in multithreaded program computation and in the model. Hence, simulation can be used for testing a parallel program with given parameters and resources.