Much of the forward thinking regarding SDN focuses on how it will enable entirely new network architectures on the virtual/logical layer. But all that software has to sit somewhere, so it’s worth it to ask: How will SDN affect physical infrastructure?
Without question, network hardware will have to change its behavior to adequately support the virtual data center. Even something as fundamental as the lowly router will see its role greatly altered once SDN takes hold. On the one hand, you’ll see legions of commodity devices stripped of nearly all their capabilities and used merely to carry out the wishes of the now all-powerful controller. On the other, you’ll see advanced devices from Cisco and Juniper that enable direct access to custom silicon through specialty APIs.
To some, this may seem like a distinction without a difference, but the fact is that the hardware underpinnings of SDN will play a large role in its ultimate utility. As Catbird’s Randal Asay noted recently, the key advantage of centralized control can be lost if there are too many device-centric configuration requirements throughout the network. Even enterprises that opt for liberal deployment of the OpenFlow protocol may not gain the full benefits of a centralized control plane, because it still relies on device-level functionality rather than a global abstraction layer.
There are already signs that the enterprise is starting to hedge its bets when it comes to router deployments. Infonetics Research recently released third-quarter numbers that saw worldwide router revenues gain a mere 2 percent, down from the 12 percent pace for the previous quarter. And analyst Matthias Machowinski noted that much of the activity is centered on high-performance devices, even at the branch, rather than the bulk commodity systems that SDN purists favor.
But is it possible that SDN would engender an entirely different kind of router altogether? One that does away with much of the internal complexity of today’s device in favor of a more flattened architecture? New routing technologies, such as the chip-to-chip silicon photonics, are already turning up in carrier-class machines like the Compass-EOS r10004 router, where they are expected to handle increasingly complex and dynamic network configurations on low-cost, small-footprint form factors. The device does away with the midplane and even the switching fabric to shuttle data with fewer bottlenecks, even as traffic scales into petabit territory. Economies of scale favor the technology at the carrier level for the moment, but it isn’t hard to imagine web-scale data centers in need of something similar as virtual, modular infrastructure takes hold.
Indeed, organizations looking beyond the next fiscal quarter and into entirely new paradigms of computing power are taking note of the need for more advanced routing technologies. The New Mexico Consortium, for instance, is checking out a new device called the Talksum Data Stream Router (TDSR), said to enable high degrees of data management, filtering and routing intelligence. The unit provides a range of real-time functions, such as data reduction, aggregation, analysis and transformation, in order to convert structured and semi-structured data into manageable event streams. In this way, the device can provide unified, coordinated information for BI and research systems, regardless of the original data type or the hardware or software platform it came from. Testing is in the proof-of-concept phase at the moment, in order to gauge the technology’s scalability and node-management acumen with the consortium’s 1,700 servers.
Of course, we’re probably getting a little ahead of ourselves here. The point of all this is that SDN is not the kind of technology that gels easily with legacy, silo-based infrastructure, so changes to hardware are in order if you intend to move networking into the logical domain. If your network equipment refresh programs are on auto-pilot, now would be a good time to re-assess your needs for the not-too-distant future.
Photo courtesy of Shutterstock.