Traffic Dynamics Online Estimation Based on Measured Autocorrelation

Abstract. Estimation of traffic demand is a major requirement in telecommunication network operation and management. As traffic level typically varies with time, online applications such as dynamic routing and dynamic capacity allocation need to accurately estimate traffic in real time to optimize network operations. Traffic mean can be estimated using known filtering methods such as moving averages or exponential smoothing. In this paper, we analyze online traffic estimation based on exponential smoothing, with focus on response and stability. Novel approaches, based on traffic arrivals autocorrelation and cumulative distribution functions, are proposed to adapt estimation parameters to varying traffic trends. Performance of proposed approaches is compared to other adaptive exponential smoothing methods found in the literature. The results show that our approach based on autocorrelation function gives the best combined response-stability performance