Network Systems Colloquium


ON THE POWER OF DIFFERENTIATED SERVICES NETWORKS

Huan Ren
Purdue University

September 6, 2PM
CS 111

Abstract

Architecting networks capable of providing scalable, efficient, and fair services to users with diverse QoS requirements is a challenging problem. The two principal issues are: design of ``good'' (a) per-hop behavior for aggregate-flow control and (b) edge control where label values are set at the boundary of the network to achieve a target QoS. In this talk, I will first introduce a theoretical framework for reasoning about optimal aggregate-flow per-hop behavior. An optimal classifier problem is solved. We show that it can be used---coupled with end-to-end adaptive label control---to facilitate scalable and efficient QoS in differentiated services networks when subject to selfish users and service providers. Next, I put the theoretical framework and analysis "to the test" by presenting a comprehensive performance evaluation study of the optimal differentiated services architecture. We use simulation to study the structural and dynamical properties of differentiated services as affected by optimal aggregate-flow per-hop control. We show the QoS provisioning prowess of differentiated services by answering the questions: What is the impact of limited bits in the TOS field on QoS provisioning? What properties does an optimal per-hop behavior satisfy? How can these properties be used to affect end-to-end QoS control (i.e., setting of DSCP values)? How efficiently are resources utilized? What are the dynamical properties of the system including stability and reachability? We show user-specified, diverse QoS requirements can be effectively facilitated over the optimal differentiated services network substrate.

Huan Ren is a Ph.D. student and research assistant at the Network Systems Lab at Purdue University. This is joint work with Kihong Park.


The Network Systems Colloquium is sponsored by the Network Systems Lab at Purdue University. For further information, please contact Kihong Park (park@cs.purdue.edu or 765-494-7821).