A team of systems and networking researchers from the Cheriton School of Computer Science has received the 2024 CNOM Test of Time Paper Award for “Dynamic Controller Provisioning in Software Defined Networks,” work that was presented originally at the IEEE/ACM/IFIP International Conference on Network and Service Management in 2013.
The CNOM Test of Time Paper Award recognizes exceptional papers published from 10 to 12 years in the past in flagship conferences and journals supported by CNOM, the IEEE Communications Society Technical Committee on Network Operation and Management. The prestigious annual award celebrates research that has been deemed outstanding and whose contents remain vibrant and useful today.
Under the direction of Professor Raouf Boutaba, graduate students Md. Faizul Bari, Arup Raton Roy, Shihabur Rahman Chowdhury, Qi Zhang, Mohamed Faten Zhani and Reaz Ahmed proposed a management framework that examines dynamic controller provisioning in software-defined networks. Since its publication in 2013, their paper has been cited almost 500 times as of June 2024 according to Google Scholar.
Left to right: Md. Faizul Bari, Arup Raton Roy, Shihabur Rahman Chowdhury, Qi Zhang, Mohamed Faten Zhani, Reaz Ahmed, Professor Raouf Boutaba
The CNOM award selection committee noted that the paper addresses the challenge of efficiently provisioning controllers in software defined networks to overcome limitations related to performance and scalability in large-scale wide-area network deployments. It is the first paper that introduces the Dynamic Controller Provisioning Problem and proposes a framework that dynamically adjusts the number and locations of controllers based on network conditions. Its contributions have paved the way for more efficient and scalable software defined network architectures, thereby influencing subsequent software defined network research and development efforts in the CNOM community and beyond.
More about this award-winning research
Software-defined networking is a new paradigm that uses network programming to configure and manage networks dynamically. By separating the control plane from the data plane and shifting the control plane to a conceptually centralized controller, software-defined networking allows network operators to implement a wide-range of network policies — such as routing, security, fault-tolerance — and to quickly deploy new network technologies.
The most common software-defined networking implementation used at the time the research team published their paper in 2013 relied on a logically centralized controller with a global view of the network. When a switch receives a new flow, it requests the controller to install appropriate forwarding rules along the desired flow path. The time required to complete this operation is known as the flow setup time. But in a large-scale wide-area network deployment, this rudimentary centralized approach has performance and scalability limits. First, it is not always possible to find an optimal placement of the controller that can ensure acceptable latencies between the controller and the switches in different geographic locations. Second, a single controller usually has a limited resource capacity and thus cannot handle large amount of flows originating from all the infrastructure switches. Here, the average flow setup time can rise significantly and degrade application and service performance
To address these limitations, proposals advocated deploying multiple controllers that work in tandem to better manage network traffic flows. But this approach introduces a new problem, namely, minimizing flow setup times by dynamically adapting the number of controllers and their locations according to demand fluctuations in the network. The researchers called this problem the Dynamic Controller Provisioning Problem.
Specifically, this problem requires enough controllers to handle the current network traffic, and their locations should ensure low switch-to-controller latencies. However, multi-controller deployment also requires regular state synchronization between the controllers to maintain a consistent view of the network. This communication overhead can be significant if the number of controllers in the network is large. Finally, as network traffic patterns and volumes at different locations can vary significantly over time, the controller placement scheme has to react to network hotspots and dynamically re-adjust the number and location of controllers. Hence, the solution to the Dynamic Controller Provisioning Problem requires finding the right trade-off between performance and overhead.
Earlier work by Heller and colleagues examined a static version of the problem where controller placement is fixed over time. These researchers analyzed the impact of the controller locations on the average and worst-case controller-to-switch propagation delay. However, a static controller placement configuration may not be suitable as network conditions can change over time.
To address this limitation, the Cheriton researchers proposed a management framework to dynamically deploy multiple controllers within a wide-area network. Specifically, they considered the dynamic version of the controller placement problem where both the numbers and locations of controllers are adjusted according to network dynamics. Their solution considered the dynamics of traffic patterns in the network, while minimizing costs for switch state collection, inter-controller synchronization, and switch-to-controller reassignment.
They formulated the Dynamic Controller Provisioning Problem mathematically as an integer linear program that considers all costs. They then proposed two heuristics that dynamically estimate the number of controllers and decide their placement to achieve the desired objectives. The effectiveness of their solution was then demonstrated using real-world traces and wide-area network topologies, with results that demonstrated that the proposed algorithms strike the right balance between the average flow setup time and inter-controller communication.
To learn more about this award-winning research upon which this article is based, please see Md. Faizul Bari, Arup Raton Roy, Shihabur Rahman Chowdhury, Qi Zhang, Mohamed Faten Zhani, Reaz Ahmed, Raouf Boutaba. Dynamic Controller Provisioning in Software Defined Networks, Proceedings of the 9th International Conference on Network and Service Management (CNSM 2013), Zurich, Switzerland, 2013, pp. 18–25.