Ronny Lam


Scott Shenker on Openflow and SDN

Ivan Pepelnjak and Brent Salisbury pointed me to this amazing lecture by Scott Shenker on the history and future of SDN.

Ivan’s conclusions are very to the point, but a bit on the “glass is half empty” side. He is a bit sceptic when it comes to SDN, but that might also be realistic. Time will tell.

My conclusions are more on the “glass is half full” side. While there were some misconceptions in the beginning of SDN in 2008(!), it has not been abandoned in 5 years time. The comparison to MPLS is great. SDN could have learned more from MPLS back then, but it still can. The abstraction of the control-plane is really helpful. As is the concept of fast “dumb” label-switching core in ASICS, with an intelligent (x86) software driven edge, controlled by a single controller.

I share Scott’s aversion to middle-boxes like firewalls, load-balancers, and the like. Not their function, but the fact that they have to be in the middle of a path and because of their intelligence are not performant enough nor are are they sharing in the control-plane. It is a good thing to move these functions to the edge of the network. Here we see a step to Network Functions Virtualization.

While I have never seen the true advantage of NFV, with this concept it really starts to make sense. Not only in the (cloud) datacenter, but also in real physical (provider) networks.

While this future may still be far away, this is something I can believe in. But then again, to get realistic again, doesn’t this again sound like LANE?