Is it too early or too late to declare 2012 the year of SDN? A few weeks ago, IBM and NEC introduced integrated technologies around OpenFlow for enterprise data centers. Last week, HP announced OpenFlow support with 16 switch models. Finally, early this week, Nicira went public with its Distributed Virtual Network Infrastructure (DVNI).
Now I'm a cynic by nature but there seems to be a fundamental transformation in progress here. Why? Because legacy data center networking equipment and operational processes are a mismatch for massive data center scale and dynamic cloud computing applications. ESG calls this imbalance data center networking discontinuity. The solution to this problem is fairly logical: Cloud platforms and server virtualization use software to turn hardware into a virtual platform. SDN, OpenFlow, DVNI, and even VMware VXLAN take the same approach.
To paraphrase the Monkees, I'm an SDN believer. Networks have to become virtual platforms that gracefully interoperate with cloud platforms like OpenStack. Provisioning new devices needs to be based upon a standard policy-based publish-and-subscribe model. Traffic engineering and security rules need drag-and-drop simplicity.
Pressing requirements, wide SDN adoption across the industry, and a new wave of engineering innovation will lead to an accelerated technology refresh cycle over the next few years. Yes, this has the potential to impact the networking status quo but I am reluctant to include hyperbolic terms like "game-changer" and "revolutionary" into my analysis for several reasons:
- Network service providers and cloud computing service vendors are feeling the effect of data center networking discontinuity most acutely so it is logical for them to experiment with cutting edge technologies. Juniper proved in the 1990s that service providers are happy to take an alternative path if they see the potential to lower operating costs or increase ARPU so Nicira's business model is spot on. By moving networking functionality to a virtual layer, Nicira gave itself a go-to-market advantage as it does not advocate a "rip-and-replace" hardware strategy (although it does turn feature-rich switches into an expensive transport layer). Still service providers feel the pain and have the motivation to move quickly.
- Enterprises are not service providers and typically don't have a team of network engineers to throw at a brand new technology. As a proof point, enterprise networking staff is finally comfortable with server virtualization although virtual switches retain the role of basic provisioning and access. Enterprises want a smooth migration path that addresses requirements and adds benefits over time without disrupting business operations.
- What about software? This market will go a few ways: A new vendor could become the VMware of virtual networking by gaining a rapidly-growing market presence. With this position, it could become the networking software nexus with published APIs, SDKs, development partners etc. Alternatively, virtual network platforms could be based upon open standards or even open source a la Linux. In lieu of either of these models however, virtual network software becomes a proprietary game. Given Cisco's market share, Cisco could turn Nexus into Windows and networking feature/functionality into Office. Regardless of the model, SDN will grow as a function of the software development community supporting it and that hasn't happened yet.
The SDN market is very exciting but remains immature. I have no doubt that enterprise data center networks circa 2017 will look very different from the device-centric, manual process-driven model of today. How we get from here to there is a bit more difficult to forecast.
You can read Jon's other blog entries at Insecure About Security.