E&CSE Research Seminar, Thusrday 26 - 04 - 2007

Abstract:

In the near future, the Internet will evolve into a multi-service network supporting a wide range of communications (video, voice as well as data). Currently, differentiated services architecture (DiffServ) is being standardized to provide the basic building blocks of this network. To ensure the quality of service demand for the new, especially time and bandwidth critical applications, differentiated services architecture introduces new functionalities into the routers for prioritization and classification of packet streams. But, the architecture does not provide an integrated, multilevel control framework for optimization of overall traffic flow, effective allocation of network resources to competing applications and maximization of resource utilization. We think that such a control system will have critical importance, but will require application of advanced control techniques. The rapidly changing network dynamics, difficulty of obtaining accurate measurements, and huge number of sensor information to be processed necessitates the application of advanced control methodologies which are able to work with imprecise, uncertain and incomplete sensor information. For example, random fluctuations of traffic flows inherently make the measurements uncertain and inaccurate.

Because of this, creation of effective network control systems becomes extremely difficult. Recent advances in fuzzy logic based multilevel control systems indicate that they are good candidates for such applications. One area in particular is promising: Type-2 fuzzy systems. We believe that they have good potential on coping with the uncertainties in network measurements as well as the non-linear nature of the differentiated services network dynamics. Type-1 fuzzy logic systems have been proven to be effective in controlling nonlinear systems of which analytical models do not exist or too difficult to obtain. But, their capabilities on processing uncertain and imprecise data have been shown to be limited. Type-2 fuzzy logic systems show good potential on remedying this, and we plan to explore especially this dimension for our work.