Study On Parallel Protocol Stack Performance Analysis

With the development of network communications that bandwidth and traffic become increasingly large, the traditional network stack is the bottleneck of system performance. So the multi-core processors and parallel computing technology has appeared, but the traditional protocol stack is not suit for it. By the use of parallel protocol stack can effectively solve the problem and improve the system performance under multi-core processors framework. So the parallel protocol stack has become a hot issue in the field of communication research and its performance has been a major concern.The performance of the parallel protocol stack was researched based on Queuing Theory in this thesis. For the network communication environments, three kinds of Queuing Models are proposed for the parallel protocol stack: Mixed Queuing Model based on M/M/1/r queuing theory, Mixed Queuing Model based on M/M/n/n+r single queuing theory and Mixed Queuing Model based on M/M/n/n+r multi queuing theory. By deriving and mathematical calculating of the queuing models, the accurate expressions of the key parameters are given. By analyzing the key parameters such as Average Waiting Time, Average Queuing Length and Loss Rate etc, we have obtained the Mixed Queuing Model based on M/M/n/n+r single queuing theory is the best queuing model for the system. It’s the solid theoretical basis for choosing the best queuing model for the parallel protocol stack system.The parallel protocol stack system is simulated by using the method of process-oriented of discrete event system. Parallel protocol stack system on packet level is simulated on the platform of Matlab. Three kinds of queuing models are dynamically simulated and the key parameters are calculated by using the statistical results. Through the analysis and contrast of the results of simulation, we can conclude that the Mixed Queuing Model based on M/M/n/n+r single queuing theory is the best queuing model for the system. The simulation results coincide with the theory analysis, which proves that the design of queuing model for the parallel protocol stack system is correct.