The network convergence is occurring at both the media and technological levels with the fast development of the digital technology. The recent development in the fiber technology of Wavelength-Division Multiplexing (WDM), promises a large amount of bandwidth in the future high-speed networks, which offers the possibility of integrating the multimedia applications together with more traditional data-oriented services within a single common network. There is a common belief that the Internet has become the dominant networking technology, but the need for multimedia communication with QoS guarantees brings forth the challenges in the current Internet traffic engineering.
Models of the traffic offered to the network or a component of the network will be critical to providing high QoS, because they are used as the input to analytical or simulation studies of resource allocation strategies [1]. In the traditional network traffic modeling, packet and connection arrivals are often assumed to be Poisson processes because such processes are mathematically tractable. However, the recent studies have shown that for both localarea and widearea network traffic, the distribution of packet inter arrivals clearly differs from exponential [2], [3], [4]. These work argues convincingly that LAN traffic is much better modeled using statistically selfsimilar processes [2], which have much different theoretical properties than Poisson processes due to the fact that there is no natural length for a ``burst'' selfsimilar traffic, that is, traffic bursts appear on a wide range of time scales.
The major challenge in the network engineering problem is that the provision of services complies with the QoS constraints while maximizing network resources. Though we can view QoS at the application or packet level, we only focus on the network performance, therefore, QoS is defined in terms of queueing delay, packet loss probability, throughput, etc. There are a number of analytical studies showing that self-similar network traffic can have a detrimental impact on network performance, including larger queueing delay and packet loss probability [8], [11], [12].
The rest of this report is structured as follows. In section 2, the basic mathematical background is introduced for the following discussions of the self-similarity. Then we present a brief survey about the current research efforts in the self-similar area in section 3. In section 4, the observations of the self-similar traffic are discussed in the perspective from the Local Area Networks (LAN) to Wide Area Networks (WAN). In section 5, the implications of the self-similarity on network performance are presented. In section 6, the tools to fill the gap between this new traffic model into the conventional network research are introduced. In section 7, the fairness property of the Earliest Deadline Scheduling with a stateless buffer management scheme (CHOKe) is also investigated on a simulation base. Finally we conclude the report and present the future work in section 8.