SDN's Next Steps: Product Integration

SDN is slowly evolving from a topic for discussion to a technology for implementation. Signs of that evolution include recent interoperability demos and hackfests. There's no shortage of SDN products from networking giants and startups alike. For instance, the Network Computing SDN Buyer's Guide has an expansive cross section of SDN hardware and software.

At present, there's clearly the whiff of prototype regarding these products--except for the OpenFlow compatible switches designed to work within a single vendor's SDN ecosystem.

Unfortunately, actual data center networks aren't homogeneous, and no one is ready to forklift upgrade to a single vendor's SDN architecture. That's the scenario the Open Networking Foundation, caretaker of the OpenFlow standard, seeks to obviate, but there's still plenty of work left in erasing the gray areas in defined standards and filling the gaps between compliant products and functioning, heterogeneous network.

However, there is evidence of SDN interoperability, at least at the demonstration phase. Perhaps the most ambitious was sponsored by Cyan Networks, where its Blue Planet SDN orchestration software served as the centerpiece of a demo at Interop Tokyo this June. The demo simulated applications running across two data centers connected via carrier networks in a hybrid cloud architecture.

The demo showed on-premise applications dynamically requesting new VMs and network resources from a cloud service using OpenStack APIs and OpenFlow. Blue Planet proxies the request, does all the work necessary to spin up new VMs, virtual interfaces and other network services, and works with servers and various network devices to allocate the required capacity.

It sounds like the classic OpenStack/OpenFlow controller example, but what set this demo apart was that it incorporated components from five other vendors: Accedian Networks' MetroNID carrier Ethernet endpoint, Arista switches, Canonical Ubuntu OpenStack platform, Overture Networks' carrier Ethernet aggregation switch and the NTT RYU open-source OpenFlow controller. These five are joined by Boundary (cloud application management) and Embrane (virtual network appliances) to form what Cyan calls its Blue Orbit SDN ecosystem. Aside from the number of vendors involved, the fact that the demo included a carrier Ethernet WAN made it particularly interesting.

If you've never heard of Cyan, you're not alone. The company just went public this spring, and, according to its S-1 registration statement, generates "substantially all" of its revenue from a line of optical switches sold to carriers, primarily Windstream, which accounted for 45% of the firm's sales last year.

The company has been building custom-developed software for its switches since day one, but decided last summer to refocus its software efforts on SDN. The result is Blue Planet, which was announced last fall and has since won about 70 customers. Cyan CTO Steve West says carriers, which he correctly notes don't just operate WANs and wireless networks but also large data centers, are likely to remain the company's biggest customer segment. The company also targets Web 2.0 firms and large financial institutions for its SDN ecosystem.

Juniper also used Interop Tokyo as a venue to announce a couple significant software automation improvements that will undoubtedly be incorporated into its SDN strategy, first announced at its Partner Conference back in January.

The highlights include support for Puppet Labs' orchestration product, enabling Puppet to manage the firm's EX, MX and QFX switches, and OpenStack integration in both Juniper's EX, QFX and QFabric switches and Contrail SDN controller. The upshot is that Juniper can now expose both physical and virtual networks as OpenStack Quantum resources.

Another recent SDN interoperability development wasn't at Interop, but at the OpenDaylight Foundation's HackFest, where 65 developers from over 25 companies came out to work on specifications and write code.

While the event focused on low-level details like OpenFlow 1.3 support, threading models in the API and ideas for building virtual tenant networks [PDF], perhaps the most important accomplishment is the sense of community and cross pollination of ideas that happens simply by getting a diverse group of engineers and developers together.

While it's far too early to declare victory, SDN proponents have reasons to be optimistic that the technology will be vendor- and hardware-agnostic, and will do for virtual network resources what Linux and OpenStack have done for compute servers.