A Long-Range Dependence Sliding Window Time-Varying Estimation Algorithm for Network Traffic

Long-range dependence (LRD) of network traffic is revealed in a dynamical evolution way. Thus, quantifying the LRD characteristics is one of the vital problems to study network behavior. Traditional LRD estimators can not give the accurate estimation under some complex conditions after seven type traditional LRD estimators are comprehensively evaluated in this paper. The main reason is that the traditional methods introduce the smoothness to traffic series in some degreedues to doing average within global domain. Consequently, some important features of network traffic such as burstiness and LRD are destroyed. A sliding window time-varying Hurst (SWTV-H) exponent estimation algorithm for LRD characteristics is proposed to improve the Hurst exponent estimating performance based on the concept of time-varying Hurst exponent induced. The SWTV-H algorithm can estimate the local Hurst exponent in some resolution ratio level, and provide a dynamic estimation of LRD trend of global behavior by shifting the local domain. The effectiveness of the SWTV-H algorithm is validated by the data of the artificial fractal Gaussian noise (fGn) series and the actual network traffic series. The results indicate that the SWTV-H algorithm is more accurate and reliable to estimate LRD characteristics compared with the traditional methods, and it has robust performance.