Correlation Based Traffic Modeling of Sub-Networks

Abstract

Internet traffic is complex and difficult to be modeled due to various traffic characteristics of different services and the application behavior. Generally, the internet is a kind of loosely-managed architecture because each sub-network can be regarded as an autonomous system (AS) and can be managed individually. In this paper, the characteristics of the sub-network traffic is analyzed from the correlation point of view. The correlation property is important for the modeling of the sub-network traffic. In addition to having the spatial correlation, the temporal traffic correlation may also exist in the sub-network due to the routine applications. It is easy to be recognized that the traffic histogram of a sub-network has the 24-hour cyclic variation phenomenon due to the daily usage behavior. The seasonal auto regressive integrated moving average (ARIMA) model is applied to characterize the above-mentioned properties of the sub-network traffic. This paper illustrates the traffic modeling of a sub-network by using ARIMA. Both small and large sampling traffic sizes are collected for modeling. The effectiveness of modeling is evaluated based on the comparison of traffic statistics and the queuing behaviors. For the comparison of traffic statistics, both the histogram and the moving average of the actual traffic collected from the sub-network and the traffic generated by the proposed model are examined for coincidence. The moving average can be regarded as the characteristics of the trend of traffic change, which is a useful heuristics for the bandwidth management. For the queuing behavior analysis, a simple queuing model test is proposed to compare the queuing behavior of the actual traffic and the traffic generated by ARIMA. The experimental results illustrate that the proposed technique can effectively capture the traffic behaviors of the sub-network and we also found that there is no significant difference in the performance of modeling between small and large sampling traffic sizes.