Scalable Performance Models for Large Scale Networks with Correlated Traffic
Project Award Date: 08-16-2001
Markovian models that capture the effects of network traffic, exhibiting long-range dependent characteristics or correlations over many different time scales, frequently suffer from the problem of not being scalable to large systems. The goal of this research is to build powerful, efficient, and scalable models that capture network behavior for varying levels of abstraction and varying finite time scales while incorporating long-range dependence and heavy tails. This entails the development of network models that can be employed at several layers of abstraction, which will allow for a full understanding of network behavior, and will use near-decomposable techniques for state reduction and time scale resolution. These models will not solely rely on steady state results, but rather on a finite horizon that would allow for online monitoring and early detection of rare events.
This project aims to study high-speed access networks, as these types of networks have been shown to have highly correlated and bursty traffic. We hope to demonstrate applications of our methods to packet level analysis of congestion control strategies over a finite window of operations; including various packet discard policies, rate and credit flow control, and random early detection protocols under realistic traffic conditions.
Primary Sponsor(s): UMKC, National Science Foundation (NSF)