SDN at the coalface: taming the “big, dirty, living network”
Last week it was the theory, this week we highlight “boots on the ground”. Yes there is some SDN activity out there, but it’s still limited and there are many heated points of contention - as we found out when we listened in on some of the sessions at ONS in Santa Clara last month. Here are the key insights from some of our ‘after session’ interviews.
The first SDN shoots are poking up above the telco legacy permafrost, but they’re still just shoots and there’s a long, long Springtime ahead. That’s what comes across when you listen carefully to a range of experts on SDN fresh from their recent SDN discussions at the Open Networking Summit (click on the URLs below for the full interviews).
It’s not that the industry doesn't understand the destination: network automation via a centralised controller. It’s just, as they used to say about telecoms standards, that there are so many SDNs to choose from. In this case there are currently about four approaches.
Recep Ozdag, Director, SDN/NFV Solutions Marketing at Cyan, says his company sits where SDN and NFV come together. Cyan provisions services with NFV and operationalises them with SDN. As regards open source and standards, Recep wonders whether standards are at all appropriate in this area but he sees two strong alternatives. The ETSI model with extensive use of Proofs of Concept (PoCs) is a good way to find out what works. The second approach is Open Source because it (eventually) becomes the standard and defines the APIs: no particular standards body needed.
So maybe the multiple SDN flavours don’t matter, as long as at least one of them eventually provides the benefits in network management, telco agility, cost reduction and so on that SDN is supposed to bring. After all, pure SDN in isolation is fairly straight-forward: it’s the transition that the industry worries about. And there are clearly different SDNs for different legacy positions.
“I don’t think it’s a secret in the industry that OpenFlow is amazing in its beauty and its terseness, but putting that into a big, dirty, living network is something entirely different,” says Carl Moberg, VP of technology at Tail-F. “It’s one thing to talk about technological excellence and how good this looks on paper,” he says. “But it’s an entirely different game if you want to put it on a product and bring it into the market.”
Being able to simulate and emulate the network will probably be important here. Mingshou Liu, president and CEO of four-year-old Taiwanese SDN specialist, EstiNet, says that, like any new technology, OpenFlow needs to be spun up (or virtually spun up) to understand its characteristics and vulnerabilities. He claims EstiNet can simulate thousands of Ver 1.3.2 and Ver 1.0.0 OpenFlow switches and run the real-world NOX, POX, Floodlight, and Ryu controllers without any modification.
Meeting the technology challenge
Here’s the challenge: how do you move SDN across that overlapping tangle of technological silos in a logical, commercially optimal way? And how do you manage the end-to-end services in the transition? Clearly in careful stages because one thing is certain: there is no way that telcos can just throw out all the existing network infrastructure and its supporting software and start from scratch.
In fact that technical diversity will be a huge challenge to the automation project, says Neela Jacques, executive director of OpenDaylight. He points out that the cost of fragmentation goes up exponentially as control is centralised. Part of OpenDaylight’s mission is to mitigate that by providing a common substrate so that applications (and real value) can be built on top.
Parts of the SDN dream have already been glimpsed. Sharon Barkai, co-founder of ConteXtream, says SDN is about digesting the lessons from Google and other over-the-top applications and taking them back into the network. His company is already mapping millions of subscriber flows to functions based on conditions. The goal, he says, is to inject optimisation, security and analytics quickly into the network using an architecture that can scale for large carriers.
For software and hardware vendors there’s still a lot to play for. Steve Garrison, VP of marketing at Pica8, says his company is betting the farm that the market will want to buy its hardware and software separately. So Pica8 has developed a hardware-agnostic operating system to work on a range of white box platforms.
Early SDN/NFV implementations are almost certain to fall short on performance. Bruce Gregory, CEO of CORSA, says he produces disruptive enabling hardware (and software) for the data plane to make up the performance gap. Again, its a theory versus reality pitch. SDN, NFV and OpenFlow are all very well in theory, says Bruce, but when telcos try to scale up their prototype networks they find the forwarding hardware can’t deliver on performance, scale and flexibility. CORSA’s DP 6400 can, he claims.