Implications of self-similarity for providing end-to-end QOS guarantees in high-speed networks: A framework of application level traffic modeling

Abstract

This paper is based on two novel research movements in the context of modeling, design, control, and management of future high-speed networks: (i) end-to-end QOS guarantees and (ii) the self-similar (fractal) nature of real-world traffic. The former pertains to building an integrated framework within which end-to-end QOS guarantees are fully supported in an efficient way while taking advantage of the statistical multiplexing. The latter concerns a salient characteristic observed in real-world traffic flow which may have critical impact on the former. However, little attention has yet been paid to determine which aspects of self-similar traffic would indeed affect control and management of future high-speed networks within an integrated framework of QOS provision end-to-end. The key objective of this paper is to address this issue by investigating (i) the distinctive characteristics of self-similar traffic from traditional Markovian traffic models, and (ii) how such characteristics would influence the three frameworks proposed in the literature. We provide evidences that the proposals considered in this paper did not properly take into account the self-similar nature of real-world traffic. As a result, a modeling methodology, called Application Level Traffic Modeling (ALTM), is proposed for more efficient QOS provision, which not only accounts for but also exploits the fractal nature of broadband traffic. An analysis of two high-quality Ethernet traces is presented supporting the ALTM. Several open research problems for further developing the ALTM are also presented.