People seem to be catching on to the fact that SDNs are going to have an impact, but I think there is still a tendency to underestimate what that might be. The Google announcement on their use of OpenFlow has prompted Light Reading to ask (rightfully) whether the fact that companies like Google are becoming a bigger piece of the network infrastructure pie so their SDN shift might be a big push on vendors. Yes it will, but what’s bigger news is that Google is a microcosm for the cloud, which is a superset of the data center. I’ve said for years now that data center network spending is driving enterprise network spending. Cloud spending is now becoming the main driver of change for operators. Google’s OpenFlow story proves that both can support OpenFlow, and OpenFlow can be a commodity switch technology.
Cisco’s buy of Truviso may be a reflection of this reality, and multi-dimensionally. First, Truviso is an analytics player that has focused on web-oriented data (though it could have other applications). That means that it could be leveraged into personalization, advertising, and other stuff that sits at the service layer. If the network is commoditizing it’s because traffic/connectivity isn’t the focus any more, and if that’s the case then focusing on what IS important (services) is smart. But second, analytics could play in the implementation of SDNs.
Cisco’s David Ward suggested in a talk at ONS that the future was a kind of feedback loop of analytics, policy, and the network. That sounds like a company getting ready to buy an analytics firm, and it also sounds like one that may be committed to engaging analytics in the SDN process. The question is how that would be done, and whether it’s a cynical ploy to try to build more legs under traditional routing or a serious attempt to build the network of the future.
The SDN notion is about having explicit central control over connectivity, replacing adaptive creation of forwarding tables. The question is how that connectivity is managed, and it has to be answered at a number of levels. You have to know what the map of the network paths is. You have to know what the state of the devices is. You have to know what connection requirements exist for applications in IPC paths and on storage-connect routes, and you have to know what the users are allowed to do. All that standard open-source controllers do is send commands to switches; you have to know what commands to send them. So how all these have-to-knows are realized and turned into switch policies is pretty important. Might this be what Cisco wants to do, what their mystery spin-out-and-back company will be doing? Possible. And remember that a Cisco guy heads the best of the cloud DevOps projects, at least in terms of the scope of its abstraction of the cloud provisioning processes. Tie that in and you have a winner.
The earnings reports of LinkedIn and Level 3 are a nice illustration of this network-evolution discussion. LinkedIn’s profits roughly doubled, which shows that social networking can be a success. Facebook’s IPO, due shortly, is another indication that’s true. Level 3 did “good” in that its loss narrowed on higher revenue. Narrower losses do not constitute a profit, much less profit doubling. The point is that you can’t expect to be a vendor increasing your sales and profits while supporting an activity that’s barely able to stay in the black. Especially when that activity is supporting a highly profitable higher layer. You need to start climbing.