On the heels of my post about Cumulous Networks last week, I had a chance to chat with Stu Bailey, founder and CTO of Infoblox, about the future of networking, and specifically OpenFlow. Stu is bullish on OpenFlow's future and the future of white-box networking in general, so there was much agreement all around. One statement he made certainly resonated: "The economics of networking are undergoing a paradigm shift like we've never seen before." I couldn't agree more.
For those who aren't up to speed on OpenFlow and white-box networking, here's a brief primer. OpenFlow essentially separates the control plane and the data plane in a network device. Traditional network devices like switches and routers make their own decisions on where Ethernet packets should travel based on rules local to the device. Those rules may be configured manually or delivered through routing protocols, but all traffic path decisions are ultimately made within the device itself.
[ Also on InfoWorld: Move over, Cisco IOS: Dawn of the industry-standard switch. | Get expert networking how-to advice from InfoWorld's Networking Deep Dive PDF special report. | Discover the top-rated IT products as rated by the InfoWorld Test Center. ]
With OpenFlow, only the data plane exists on the switch itself, and all control and pathing decisions are communicated to the device from a central controller. If the device receives a packet for which it has no pathing or flow information, it sends the packet to the controller for inspection, and the controller determines where that packet should be sent. The controller can then add a flow entry to the switch for handling future packets of the same type.
The instructions sent to the switch could be anything. They could instruct the device to drop the packets and all future packets of the same type, essentially creating a firewall. They could instruct the device to modify the headers and pass the packets to a specific port, while a new packet stream to the same destination is rewritten with a different header and passed to a different port, creating a load balancer. They could tell the switch to throttle or prioritize the packet flow for QoS purposes. All of this can happen on a network switch that essentially has no operating system or configuration other than the knowledge of how to contact the controller.
This, naturally, turns traditional networking on its head, and it's becoming clearer just how necessary that really is. Traditional networking is having difficulty dealing with modern computing scenarios, especially with virtualization, and many new concepts and ideas are being applied to that traditional model to deal with those issues. Virtualization vendors are taking networking decisions away from the network and handling them within hypervisors. Data-center-bridging technologies have popped up to deal with the fact that virtual machines can traverse physical buildings quite easily now, whereas the traditional IP networks they're attached to aren't so mobile. All of this becomes moot using OpenFlow, because we can program the network as a whole, centrally, versus managing compartmentalized networks adapted to physical locations.