Event Type:

Probability Seminar

Date/Time:

Tuesday, November 27, 2012 - 07:00 to 08:00

Location:

Kidder 278

Guest Speaker:

Institution:

EECS, Oregon State University

Abstract:

We present a new analytical framework for designing network protocols and application policies under non-stationary environments. In order to cope with non-stationary environments, e.g., fluctuating nature of traffic volume or fading in wireless channels, it is preferable to have a network protocol that adapts and quickly achieves a given objective. In this context, we use the concept of mixing time in Markov chain theory to study the convergence rates of different protocol designs. As an application of our framework, we propose a convex optimization approach for obtaining the robust queuing policy that drives any initial distribution of packets in the queue to the target distribution in the fastest time. We then augment the proposed framework to obtain a queuing policy that produces an epsilon-approximation to the target distribution with even faster convergence time. The augmented framework is useful in dynamic settings where the traffic statistics changes frequently, and thus fast adaptation is preferable. Both simulation and theoretical results are provided to verify our approach.