Making SDN’s REALLY Work

The challenges with coming to terms with technical concepts in our Age of Hype is embodied in a story about SDN that appears in Network World.  The story lists five drivers of SDN transition, which include mobility, the cloud, consumerization of IT, traffic patterns in data centers, and agile service delivery.  The challenge is that in a direct sense these challenges work against what most would say is “SDN” and not toward it.

Things that are attributes of service usage need a service as a foundation, and as I’ve noted the purist form of SDN, embodied in OpenFlow, is about per-device forwarding control and not collective service behavior.  If an application or user wants a “service” from an OpenFlow network, the way that network resources function collectively to move traffic has to be known by the central intelligence guiding the forwarding process.  In short, there has to be a service, and this was the point that Cisco was making at Live; if you have to support current services with SDN principles it will be easier to work through current protocols.  Cisco ONE proposes, in effect, to apply application intelligence but not from a single central point.

The example of Google’s OpenFlow application is cited by most SDN advocates, including myself, but it’s important to recognize that what Google did was to embed OpenFlow inside a service network.  That means that OpenFlow addressed not the totality of the service needs but the needs of a collection of devices embedded in the service.  Even that wasn’t a particularly easy job; at the boundary point between IP and OpenFlow you have to satisfy the “language” or control behavior of the IP network or the OpenFlow part won’t work in harmony.

The point here is that this whole issue of OpenFlow and SDNs is a lot more complicated than most people think it is, and the most complicated part of all is just how SDN principles can be harmonized with the current Ethernet and IP networks, for which all or applications are written and all our devices wired.  The notion of “embedded SDN” makes a lot of sense because it simplifies the problem by containing it to somewhere inside the network where “services” really aren’t visible except as control exchanges among devices.

Wall Street upgraded Ciena today, and even here we see an issue.  Ciena embraces the notion of SDN control over optics, which could be the key to creating an OTN that functions inside an IP or Ethernet core but doesn’t require a lot of IP/Ethernet logic be duplicated in order to work.  The Credit Suisse research report makes some good points about the trends in optics, and the fact that operators would at this point rather spend on optical capacity than adding electrical features to manage bandwidth that’s eroding in price by 50% per year.  The thing is, all of those benefits would be maximized by applying the SDN OpenFlow and Google model to optical cores.  What OpenFlow does here is to support forwarding systems that can look like virtual devices, hiding all the interior structure and features.  Think of it as everyone peering with a giant router, making everything one hop away.  Yes, there are issues of harmonizing with routing protocols, but that’s where OpenFlow can shine.  If, of course, everyone works on the problem.  Which happens only if somebody stands up and says we need to.

Cisco, meanwhile, is working on architectures and at the same time getting ready to take advantage of market conditions to steal some market share.  In their rhetoric they’ve identified two prime competitive targets, Huawei and Juniper.  In the case of Huawei Cisco’s moves are defensive; the Chinese giant has clear price/cost advantages and Cisco doesn’t want the market to sink into pushing featureless bits or Huawei will win.  In the case of Juniper, Cisco’s in predator mode.  They believe, following another Wall Street theme, that Juniper simply cannot keep up at this point.  Much of what Cisco is delivering in its architectures announcement were talked about by Juniper, as long as four years ago, but that have stalled in delivery.  One concept, “architectures”, positions Cisco.


Leave a Reply