The Threshold Crossing Method On High-speed Network Control

Providing the quality of service (QoS) has become an important research field with the development of network communications. Network congestiondetection and control is one of the major approaches in providing QoS.Traditionally, many complicated stochastic models, which discribe the short range dependence, were introduced to model the network traffic, such as Poisson process, the fluid-flow model, the packet-train model, the Markov-modulated Poisson process, the batch-arrival Markov process, and so on. Through the study of network traffic, researchers showed that the network traffic actually has self-similarity statistically and long range dependence. Recently FBM presented by Duffield and FARIMA provide a better way in modeling the network traffic, traffic prediction, and admission control. However, these models are too complicated to be used practically, which makes it difficult to conduct network traffic control based on network traffic prediction due to the computational complexity.Optimizing the bandwidth allocation and admission control are often used to ease the network traffic congestion and thus provide quality of service. To do so, we are more interested in the network traffics that exceed some threshold value, and pay less attention to the network traffic below the threshold value since such traffic does not cause much difficulty in network traffic control. We propose a feasible threshold-crossing method that does not take the statistic distribution of the network traffic into account, and avoids the mathematical difficulty in processing previous models. The selection of the threshold value is critical here for the threshold-crossing method. Through the study of the threshold selection, we can estimate the proportion of the arrival traffic that exceeds the threshold value, and its impact on network congestion. This can be used to estimate the buffer size needed in storing unprocessed packets in providing the quality of service.Specifically, we design an energy function of the threshold using variance analysis for high-speed network traffic. We conduct extensive analysis of the data pAugTL1s (3000 data items) and pAugTL100 (30,000 data items) from Bellcore, we obtain some regularity in such traffics, which provides the basis for threshold selection. Based on threshold-crossing method, we studied the feasibility for network control. Our measurement and analysis show that it is possible to compute the threshold value in a short time. And then we discussed how to allocate dynamically buffers by threshold crossing method, and gave the arithmetic.