My Latest Thoughts on OpenFlow

SDN event at Stanford just the beginning

I've been thinking about OpenFlow for a while as it relates to data center networking. What's the use case? What's the enterprise play? What will OpenFlow and SDN do to the networking market? I've seen the "plugfests" and know that OpenFlow works. I spoke with Broadcom at Interop and understand how it is being supported in silicon. I also saw a very impressive demonstration from NEC which went beyond a proof-of-concept to emulate an enterprise network complete with automated tools for provisioning and operations. Just this week I had a very good conversation with Big Switch about SDN today and its future. So here's a few of my latest thoughts: 1. I think "Software Defined Networking" is the best way to describe this category. While I am bullish on OpenFlow, it is really a network protocol and other technical details around SDN implementation. To me, the really important thing going on is the use of software to align network control with application and business needs. 2. SDN is already a no-brainer for cloud computing. If we really want to move workloads around dynamically, we can't be bound by networking constraints like IP addresses ranges, sub-nets, and VLANs. SDN can address this limitation. 3. SDN will become a no-brainer in the largest of the large enterprises. Same thing applies here for large technically-sophisticated shops building private clouds, but SDN will also be required in very large networks with massive server-to-server traffic supporting SOA applications and Hadoop backends. Even smaller networks can benefit from SDN's ability to help automate network provisioning, segment networks for security/compliance and ease network management headaches. 4. SDN may align applications and networking requirements. Ultimately we want to provision applications at the drop of a hat without jumping through networking hoops. SDN holds promise here. To me, the big remaining questions around SDN are around architecture. Can an external controller really scale to provide control path guidance for a massive network? How do we an external control path network that maintains real-time integrity and availability? These are familiar computer science problems that have been addressed in the past so I'm confident that SDN proponents will figure this out. What's more we can agree on some standards and then let vendors like Arista, Big Switch, Cisco, HP, Juniper, and NEC provide additional functionality and develop killer software. For a grey-hair like me, SDN is a lot of fun. It reminds me of other collective and open source efforts in the past like OSF, Linux, Apache, and Java where people in the industry showed real passion. The thing about SDN however is that it goes beyond technology alone and makes real business sense. We need a way to make application provisioning, workload migration, and cloud computing as easy as possible. SDN's vision of a programmable, flexible, and customizable network really could help us get to these goals sooner rather than later.

FREE Download: Get the Spring 2019 digital issue of CSO magazine today!