The Open Networking Summit met all the expectations!
Representatives from academia, industry, investors, specialized media, gathered together and discussed their visions and interpretations on OpenFlow/SDN.
Great talks and a general consensus that SDN is a disruptive approach to networking, with OpenFlow being "just" the catalyzing technology to abstract the forwarding hardware by means of a pragmatic instruction set.
Another fundamental issue that was discussed is that adoption and usage of OpenFlow/SDN will not be an "all or nothing" choice, that is, an hybrid networking environment will be the rule (and I would argue that not only during a transition phase). It will be very much deployment scenario specific but a common use case shall be that a number of networking features / services will be offered by an OpenFlow-enabled control plane while the remainder of the traffic will be handled by the traditional (hardware-embedded) control plane. That is, there may be traffic coming from the same end-host (ie., from a certain application/service) that will be ruled by an OpenFlow control application.
The key here is dealing with and designing for hybridism.
Indeed, RouteFlow is about allowing hybrid SDN network but taking a different approach by moving the legacy IP routing stack to the controller application. By talking to attendees during the event, demo visitors valued RouteFlow as a means to migrate to SDN by allowing the flexible deployment of traditional IP networks on top of OpenFlow infrastructure. This is certainly a first step and a use case for RouteFlow. Once that we can show in scale that this is viable approach, we can start innovating by optimizing the IP routing strategy (e.g., separating route state distribution from physical topology maintenance) or simplifying the network design by aggregating or multiplexing physical switches as convenient.
Back to our demo at ONS. The booth was packed of visitors most of the time, raising interesting questions and ideas for new deployments and use cases that we hope to work on and in collaboration. Very interesting was hearing how third parties would explain RouteFlow to each other. Also nice was realizing that on team of Stephen Stuart (Google) was "playing" with RouteFlow to provide a "glue" between open routing stacks and OpenFlow. We learned about the Open Source Routing Project and their goal of turning Quagga into production quality, and there is a plan to experiment with RouteFlow over the NDDI infrastructure in collaboration with Indiana University.
Talking about IU with whom we shared the booth at ONS, Chris Small (right in the picture below) demostrated RouteFlow doing BGP peering with two Juniper boxes while running a four OSPF router network carved out of a single Pronto switch by virtualizing it into multiple datapaths. Cool! First demo with real switches! More demos to come next week in the OFELIA/CHANGE summer school and the week after in the Super Computing 2011 SCinet Research Sandbox in Seattle!!