12:45 PM, Thursday, November 8th, 2001
Room 041, Building 420 Note location change for today!

Effairness: A Metric for Congestion Control Evaluation in Dynamic Networks

Sergey Gorinsky
UT Austin

About the talk:

The Internet is a dynamic environment with a diverse and quickly changing mix of network sessions; some of them - e.g., short Web transfers or lasting multimedia streams - are interested in short-term efficiency and fairness of data delivery. In this talk, we examine the problem of congestion control evaluation in dynamic networks. After highlighting the deficiencies of the existing metrics for congestion control performance, we identify a fundamental reason for these inadequacies - the existing metrics are defined with respect to ideal allocations that do not represent short-term efficiency and fairness of network usage in dynamic environments. We proceed by introducing an effair allocation, a dynamic ideal allocation that specifies optimal efficiency and fairness on every timescale. This effair allocation possesses other desirable properties including its ability to characterize an ideal allocation in networks with multicast sessions as well as its dependence on the communication needs and capabilities of applications. We design an algorithm that accounts for network delays and computes the effair allocation as a series of static ideal allocations. Using the effair allocation as a foundation, we define a new metric of effairness that shows how closely the actual delivery time for an application-specific chunk of data matches its delivery time under the effair allocation.

About the speaker:

Sergey Gorinsky is a doctoral student in Computer Science at the University of Texas at Austin. Native of the Soviet Union, he holds a degree of Engineer in Computer Engineering from Moscow Institute of Electronic Technology and an M.S. degree in Computer Science from the UT Austin. In 1998 and 1999, he worked at Lucent Bell Labs and AT&T Labs Research.

Sergey has been investigating various aspects of resource allocation in computer networks, including real-time scheduling and admission control in integrated-services networks. Currently, his research focus is congestion control in the Internet.