DOI: 10.17587/prin.9.59-68

Comparison of Sequential and Parallel Software Implementations of the Hybrid Fluid Model of Information Flows for Computer Networks with Complex Topology

D. A. Basavin¹, basavind@gmail.com, S. V. Porshnev², sergey_porshnev@mail.ru, D. A. Petrosov¹, scorpionss2002@mail.ru, ¹Belgorod State Agricultural University named after V. Gorin, Mayskiy, Belgorod region, 308503, Russian Federation ²Ural Federal University named after the first President of Russia B. N. Yeltsin, Ekaterinburg, 620002, Russian Federation

Corresponding author: Basavin Dmitry A., Assistant, Belgorod State Agricultural University named after V. Gorin, Mayskiy, Belgorod Region, 308503 Russian Federation, E-mail: basavind@gmail.com

The article discusses the analysis and comparison of software implementations of the hybrid fluid model (HFM) of information flows in computer networks with complex topologies. The sequential version implemented on the central processing unit (CPU) and the parallel version implemented on the graphics processing unit (GPU) using general-purpose computing for graphics processing units (GPGPU) technology are considered. During the analysis of the GPU based HFM implementation, it was verified against the CPU based implementation on a series of experiments. The results of the experiments confirmed the convergence of the results obtained by both versions and are presented in the article. A number of quantitative information flows characteristics were compared graphically and numerically to verify the results. Thus characteristics were the minimum, average and maximum input and output aggregated flow rates on routers. A number of experiments were performed for performance estimation and acceleration of the parallel HFM relatively to sequential. The parameters of the experiments were selected to maximize the use of available computing resources. The size of the simulated network for each experiment increased exponentially. For each experiment, measurements of computation time for sequential and parallel implementations of HFM were carried out, after which they compared graphically and numerically. During the comparison of software implementations, the hypothesis was confirmed by the advisability of using GPGPU technology to accelerate the calculations of HFM. It clearly demonstrated that the parallel software implementation makes sense to apply to the number of information flows >2^6.

Keywords: Internet traffic, computer networks, parallel hybrid fluid model, modeling, GPGPU

pp. 59–68

For citation:
Basavin D. A., Porshnev S. V., Petrosov D. A. Comparison of Sequential and Parallel Software Implementations of the Hybrid Fluid Model of Information Flows for Computer Networks with Complex Topology, Programmnaya Ingeneria, 2018, vol. 9, no. 2, pp. 59—68.