event

PhD Defense by Hyojoon Kim

Primary tabs

Ph.D. Dissertation Defense Announcement

Title: Facilitating Dynamic Network Control with Software-Defined Networking

Hyojoon Kim
School of Computer Science
College of Computing
Georgia Institute of Technology

Date: Friday, May 1, 2015
Time: 2:00 PM - 4:00 PM ET
Location: KACB 3402

Committee:
Dr. Nick Feamster (Advisor, School of Computer Science, Georgia Institute of Technology)
Dr. Ellen Zegura (School of Computer Science, Georgia Institute of Technology)
Dr. Mostafa Ammar  (School of Computer Science, Georgia Institute of Technology)
Dr. Nate Foster (Department of Computer Science, Cornell University)
Dr. Yoshio Turner (Banyan)


Abstract:
Network management is complex and error-prone; this complexity and brittleness has several causes. First, much of today’s network management process remains low-level and manual: operators must update router and switch configuration files to change the network’s forwarding behavior while configuration language is unintuitive and hard to understand. Second, the configuration is distributed all over the network instead of residing in one location, which makes it harder to manage. Third, network conditions are dynamic. Traffic patterns change, hosts arrive and depart, topologies change, intrusions occur, and so forth, thus the network’s configuration needs continuous updates by operators. Yet, we understand very little about the nature of network configuration and its changes. We also lack alternate solutions that can help operators to express today’s complex network policies with less misconfigurations.

This dissertation analyzes traditional network management methods to better understand the problem in network configuration, and presents better network management solutions that help operators to configure and program their network in a concise, intuitive, and less error-prone way. First, we analyze over five years of historical network configuration files from two big campus networks and show that a network can experience a lot of changes; number of lines that change in all configuration files ranges from 200,000 to 800,000 lines per year. Based on our findings, this dissertation presents two distinct solutions that are both based on Software-Defined Networking (SDN). Kinetic is a domain specific language and network control platform that enables operators to write programs that can automatically reconfigure the network in face on arbitrary network events. Kinetic also automatically verifies the correctness of these control programs with respect to user- specified temporal properties. Coronet is an SDN-based service that provides automated recovery in face of an unexpected but common network event: data-plane failures such as switch and link failures. Coronet provides other SDN control applications with the abstraction of a reliable virtual topology and achieves fast failover by updating small number of table entries upon failure detection rather than the much larger number of individual flows that are assigned to those table entries.

Our user study of Kinetic with several hundred network operators demonstrates that Kinetic is intuitive and usable, and our performance evaluation shows that realistic Kinetic programs scale well as the number of policies and the size of the network increase. Evaluation of Coronet against four Mininet-emulated network topologies demonstrates that Coronet can provide fast failover (maximum less than two seconds).

Status

  • Workflow Status:Published
  • Created By:Tatianna Richardson
  • Created:04/17/2015
  • Modified By:Fletcher Moore
  • Modified:10/07/2016

Categories

Target Audience