Farima Model-based Network Traffic Modeling And Prediction

Since self-similarity challenges traditional network models, which have biases to real network traffic. It leads difficulty to network traffic modeling and analysis. It's imperative to research and study new models. Distinct to other common models, FARIMA model is able to describe both short dependence and long dependence structure, is particularly suitable for self-similar network traffic modeling and prediction.A simplified FARIMA model is proposed after theory analysis, simulation and result analysis. It can be used in network bandwidth allocation and network traffic flow control. The primary studies are list bellow.1. A simplified FARIMA model is proposed. FARIMA process can be divided to fractional difference process and ARMA process, through mathematical transform. Accordingly, we estimate parameters of FARIMA process by these tow steps above. The impact of parameters to the result is studied. Considered the predictive value of ARMA process tends to mean value at negative exponential speed, we proposed a simplified model: ARMA predictive value can be replaced by mean value directly, in long term prediction. By contrast the performance of FARIMA model and the simplified one, the full FARIMA model's prediction is more accurate than the simplified one in short term; while in the long term, the simplified FARIMA model's performance can approach the full model's. Thereby, the simplified FARIMA model is valuable and useful in long term prediction. It can reduce about 30% computational costs.2. Dynamic network bandwidth allocation. We established a procedure for dynamic bandwidth allocation scheme in frame of upper probability limit. The relationship of buffer size, bandwidth allocation scheme, queue length and packets loss rate, is obtained after simulation and result analysis. In infinite buffer size, dynamic bandwidth allocation scheme got shorter queue length than fixed bandwidth allocation scheme. In finite buffer size, dynamic bandwidth allocation scheme got less packets loss rate. Accordingly buffer size can be saved in dynamic bandwidth allocation scheme.3. An improved flow control protocol is proposed based by introducing FARIMA predictor in WSN. We design the PBESRT procedure, present the state transform and report frequency adjustment schedules. Simulation compared PBESRT with the original ESRT. Then the performance of each protocol is studied at deferent Hurst parameters value. The result shows that PBESRT converges to state OOR with less time intervals, starting from some special states. More importantly, with different Hurst parameters the network with PBESRT protocol runs more time intervals at OOR state. It improves the performance from 1.6% to 8.8%. Consequently, Prediction-based ESRT protocol makes sure the network attains required event reliability with less energy consumption, prolongs the existence period of wireless sensor networks.