Network Flow Scheduling with Zero in-Network Buffering and Non-zero Reconfiguration Delay

11:00 AM - 12:00 NOON
Atkinson Hall, Room 4004

Prof. Young-Han Kim
Tara Javidi, Associate Professor,ECE, UC San Diego

In-network queuing and distributed operation have been essential attributes of the classical TCP-IP networking ensuring scalability and fast growth. However, this architecture has come at the cost of excessive delay and tardy flow completion times. In recent years, and with the growth of inter-data center networking, new architectures are proposed which drastically depart from this classical approach and avoid in-network queuing all together. In this new class of network solutions, often serving a densely packed set of stationary end hosts, the network can be managed by a single entity/arbiter/controller, allowing for fine-grained management and scheduling of flows across the network. By separating the control and data planes, the proposed hybrid architecture avoids in-network queuing and can potentially result in significant reduction in latency at the cost of a fine-grained monitoring and control of the network traffic.

The overall objective of this talk is to develop a scheduling framework, from first principles, which utilizes this newly popularized networking architecture of separated control and data planes to increase efficiency and significantly reduce latency. We will show that the critical technical challenge is to design end-end scheduling algorithms that account for non-negligible reconfiguration as well as monitoring delays. Furthermore, the design has to optimize the computational complexity of the scheduling algorithm as well as the cost of monitoring across the network. In this context, this talk considers generalized yet simple analytic models at the intersection of stochastic control and information theory which capture the most significant attributes of the problem. Furthermore, we summarize our recent theoretical results on the design of adaptive schedulers as a promising building block of the efficient low-latency solutions.

Based on joint work with Chang-Heng Wang.

Speaker Bio: 

Tara Javidi studied electrical engineering at Sharif University of Technology, Tehran, Iran from 1992 to 1996. She received her MS degrees in electrical engineering (systems) and in applied mathematics (stochastic analysis) from the University of Michigan, Ann Arbor, in 1998 and 1999, respectively. She received a Ph.D. in electrical engineering and computer science from the University of Michigan, Ann Arbor, in 2002. From 2002 to 2004, Tara was an assistant professor at the Electrical Engineering Department, University of Washington, Seattle. In 2005, she joined the University of California, San Diego, where she is currently an associate professor of electrical and computer engineering. Tara Javidi’s research interests are in communication networks, stochastic control theory, information theory with feedback, and wireless communications.

Speaker Photo: 
Event Date: 
Friday, November 20, 2015