Journal "Software Engineering"
a journal on theoretical and applied science and technology
Issue N6 2018 year
The article discusses strategies of load balancing in a heterogeneous computing environment that includes the central processing unit (CPU) and the graphics processing unit (GPU). The authors used these strategies during development of a hybrid fluid model (HFM) of information flows in modern computer networks with complex topologies. The HFM represents system of ordinary differential equations (ODE) from math point of view and a balance equation of informational flows that comes in and out from a corresponding node of a modeling computer network from physical point of view. Due to the lack of analytical solutions for systems of ODE, which are included in the HFM, it is necessary to develop appropriate software tools that allow making numerical solutions of ODE system in acceptable time. It was noticed that a computation time of modeling of information flows characteristics in high-speed computer networks significantly depends on distribution of a computational load between components of the heterogeneous architecture. This assumption was formed during development and using of the parallel HFM software implementation based on the general-purpose computing for graphics processing units (GPGPU) technology. Due to this fact there is a need to choose a data exchange strategy between the random access memory (RAM) and the GPU. It should consider HFM computing algorithms details and provide balanced distribution of the computational load between components of the heterogeneous computing system. This paper describes the results of experimental researches of data exchange strategies.