The Internet2 conference this week was a smashing success. Reflecting on the audience’s comments and interest in our advanced development SDN (Software-Defined Networking) infrastructure testbed leads me to elaborate on it a bit further.

SDN was conceived as a method to control and virtualize elements inside data centers, not between them. But the testbed at Marist College in Poughkeepsie, N.Y., stretches the old bounds of the concept.

It utilizes data center switching, server and storage technology from IBM and long-distance optical networking equipment from ADVA Optical Networking among three simulated data centers. It all operates under the control of open-source SDN controllers from FloodLight and eventually the OpenDaylight Project. The SDN controller marries the network’s packet and optical domains by leveraging the OpenFlow protocol, modified with certain key extensions to deal with the analog nature of the optical network. These include optical-specific constraints in areas such as sequential lightpath setup/teardown, optical power balancing, switching and wavelength continuity.

Most SDN testbeds focus on the local area network (LAN) inside the walls of the data center. The Marist testbed, however, is different in that it extends SDN functionality to orchestrate both the LAN and the wide-area network (WAN). The testbed shows how resources can be cost-effectively pooled among data centers by automating wavelength provisioning, commissioning and assignment, while equipping the network to dynamically respond to on-the-fly changes in application workload and traffic patterns.

These are timely capabilities for Internet2 communities and telecommunications and cloud service providers. Given the rise of on-demand resource and application models in the cloud, such network operators are dealing increasingly with dynamic workloads and traffic patterns that can change quickly in response to application demands. Network protocols that were not designed to address such requirements are making service provisioning and network scalability more cumbersome and costly. SDN, on the other hand, leverages a centralized, logical view of the network that can be easily manipulated via software to implement complex networking rules. The resulting benefits include support for multi-vendor environments, more granular network control (at session, user and device levels), improved automation and management, accelerated service deployments and unprecedented scalability and flexibility at lower cost.

Let’s look more closely at how the Marist SDN testbed works.

In the first step of testing, reconfiguration is initiated from the SDN controller, which, after performing necessary calculations, programs the wavelength division multiplexing (WDM) system and OpenFlow switches to set up a flow between two geographically dispersed data centers. Triggered by preset scheduling or through a web portal, a path and bandwidth is released to the given application(s).

In the second step of testing, reconfiguration is initiated from the application itself. In this scenario, the application requests a path and bandwidth from the SDN controller, which then programs the WDM systems and switches and releases bandwidth accordingly.

When completed, the “application-aware” SDN controller releases the bandwidth back to the LAN/WAN pool. 

Various use cases can be demonstrated in the testbed.

For example, one growing use case is workload balancing. In the Marist scenarios, vmWare vMotion could be used to move virtual machines and vStorage to move the associated files both in and between data centers for more effective use of resources. In this scenario, a system (server or storage) could be reaching maximum capacity, which would trigger the need to move virtual machines to another less utilized server and storage in a remote cloud data center. The application could request two 10 Gbit/s wavelengths between data centers A and B. The SDN controller would check to see what switches, optical gear and other network resources are available and then provide bandwidth back to the applications. Once virtual machines and files have been moved from Data Center A to Data Center B, the application would give the 10 Gbit/s wavelengths back to the metro pool, and the same process can be demonstrated between Data Center B to Data Center C. (See Figure B.)

Another possibility is flooding a link with virtual machines, storage and/or video traffic, causing another 10 Gbit/s wavelength to be activated when capacity reaches a given threshold—say, 90 percent—on the first link. This is much more efficient and cost effective than having a second 10 Gbit/s link just for backup and only getting one half the bandwidth. This particular 10 Gbit/s link comes from a “pool” that can be shared with other 10 Gbit/s links, providing “wavelengths on demand.” (See Figure below)

The testbed demonstrates the basic functionality of an SDN-enabled service provider network and illustrates the compelling cost efficiencies and revenue opportunities enabled by new levels of dynamic provisioning and automation across network layers.

The potential real-world applications that could be enabled by the capabilities demonstrated in the Marist testbed are promising. More powerful services for disaster recovery and business continuity could be supported among multiple data centers separated by distances up to 300 kilometers. In the event of a natural disaster such as a hurricane, it would be beneficial if the inter-site network could quickly be reconfigured to facilitate transferring all mission-critical data and applications outside the affected region.

“Big Data” analytics could be supported among several data centers with a prescribed traffic pattern and network bandwidth allocation that changes based on workload. Fiber-optic links could be combined into a larger virtual pipe to accommodate a momentarily heavy flow of traffic—with the links automatically separating under supervision of the controller when the rush is over. Or, a network administrator could isolate certain long-distance flows in the network for differentiated, on-demand “express lanes” for time-sensitive voice, video and data traffic.

SDN across network layers via OpenFlow, as demonstrated in the Marist testbed, stands to transform networks from infrastructure to business-critical service-delivery platforms. Network control and management is simplified, and more agile deployment of network services is enabled. For Internet2 communities and telecommunications and cloud service providers, such new functionality translates into significant cost efficiencies and the ability to seamlessly adapt to rapidly changing needs—benefits that address the most pressing challenges that these network operators today face.