Environment for Tuning Parameters of a Multithreaded Program Developed Using a Dependency Graph

Abstract

With the advent of multi-core central processors, multithreading has become the most widespread practice for improving program execution performance. However, the development of a multithreaded program remains a rather complex process. To simplify this process and enhance the performance of the resulting program, various methods for managing thread-based execution are often employed.

One such method is the method of managing the execution of tasks of a multithreaded program according to a given dependency graph. This method significantly reduces the resource intensity of program development and increases program performance by employing a lockless approach to multithreaded programming.

Nevertheless, the challenge of efficient utilization of computational resources remains relevant and can only be addressed through the careful design of parallel computations. In particular, identifying the configuration parameters for a multithreaded program that ensure optimal resource utilization is a resource-intensive and complex task, even for highly qualified specialists.

This study examines existing approaches to tuning the parameters of multithreaded programs to achieve the most efficient execution. It proposes the use of an environment for tuning multithreaded program parameters based on the method of managing the execution of tasks of a multithreaded program according to a given dependency graph. The accuracy of the resource efficiency metrics obtained through this environment was experimentally validated. A practical example demonstrates the application of the environment in the development of a multithreaded program. The use of the environment also facilitates the configuration process of multithreaded program parameters.