SDN and NFV: AT&T's root-and-branch approach
28 May 2014 | Richard Irving
To many, AT&T epitomises all that is wrong with today’s top-tier telcos: cautious, conservative and conformist. Not anymore. The top brass are backing an audacious root-and-branch network revamp.
You may be forgiven for thinking that the hyperbole surrounding Software Defined Networking (SDN) and Network Functions Virtualisation (NFV) – for, yes, we must now mention both these technologies in the same breath – could not get any more feverish.
But AT&T managed to ratchet up the excitement to a whole new level in March, when it set out a detailed plan to rip out its current network in favour of a bleeding-edge, NFV-powered SDN replacement. That is quite a mouthful even in the acronym-riddled world of telecoms, so let me explain.
The concept behind SDN is to take all the over-engineered intelligence that sits in hardware out in the field, such as routers and switches, and migrate it to a new breed of super-computer that will nestle above the network, optimising traffic flows and making hitherto unimaginable applications a possibility at, quite literally, the touch of a button. The premise, nay, the promise of NFV, is that it will provide the programs that will spawn those products and services – applications that will be limited only by the imaginations of those software specialists writing the code.
So how has AT&T managed to whip up a market already frothing with anticipation into a new frenzy?
For better or worse, the clue lies in the name. This is AT&T we are talking about – an operator with a risk-averse nature that typically deals in decade-long development cycles; a carrier steeped in a corporate culture that, critics say, has changed little since the days of the Ma Bells; a company where you still wear a suit and tie and use courtesy titles when addressing your peers. And yet, alone among a sea of competitors looking to stake their SDN claim, AT&T has set out a detailed roadmap that challenges just about every preconception you might have of the US giant’s strategy and guarantees to turn every aspect of the way it does business on its head.
There will be no compromises: no cosy cabals of longstanding equipment vendors quietly driving the process according to their own agendas. Indeed, there will not even be any provisioning for a network overlay to help with the transformation.
There will, by contrast, be root-and-branch change: changes to the way the company designs new infrastructure; changes to the way that infrastructure is commissioned and purchased; and changes to the way the network is subsequently managed. As John Donovan, AT&T’s EVP of IT and infrastructure said at the launch, “We’re moving out, and we’re moving out in a big way”.
In Donovan, a former championship boxer and keen ice hockey player, the board of AT&T has found a pivotal conduit to drag the lumbering telco kicking and screaming into the 21st century. When he joined the carrier as CTO back in 2008, Donovan’s brief was to take on the challenges that AT&T felt it could no longer tackle intuitively. He was tasked with instilling a Silicon Valley mindset into an R&D effort that had been plunged into paralysis by the march of the OTT segment. His bosses wanted him to seek inspiration from further afield than the operator’s very own innovation arm, AT&T Labs – once a behemoth in network design and the birthplace of, let’s not forget, the communications satellite.
The big shake-up
And Donovan has delivered. In the last 3 years, he has met with more than 1,200 startups, searching for the technologies that will change the way the telecoms world works for good. Now, he is making his move. Asked why he holds so much store by SDN, Donovan is unequivocal.
“There is no army on this earth that can hold back an economic principle whose time has come,” he says, adding that today’s telecoms infrastructure is not fit for the modern digital age.
“This is an urgent call to action. Are today’s networks easy to scale? No. Are they cheap to run? No. Are they adaptable? No. Do we need to transform them for the future? Absolutely… What I’m talking about here is a radical reshape of the entire wide-area network,” he says.
Enter the “user-defined network cloud”, a heady mix of SDN and NFV technologies that promise to do for AT&T’s network what the cloud did for computer processing and storage. The solution, also known as Domain 2.0, is built around 4 key principles: firstly, the network will support open APIs that will allow customers to manipulate and consume services however they want; second, it will be simple in design, lacking any of the complex hardware that currently needs rebooting every time tweaks are made to the architecture; thirdly, it will be scalable, capable of coping with the kind of ongoing growth that has seen traffic volumes at AT&T surge 50,000% – yes, 50,000% – in the last 7 years; and finally, the network will be secure, safeguarding the integrity of its “brains” with cutting-edge systems.
What sets Donovan’s ambitions apart from the whimsies of an IT fantasist is that he has already got top-level approval to turn the culture at AT&T on its head in order realise this vision.
“We’re changing everything”, he says. And he really does mean everything, from the way the network will be designed using very clever software (rather than very expensive hardware), right through to the way it will be built, and by whom. AT&T has already earmarked four equipment vendors with whom it plans to work closely on the build-out, including Ericsson, Metaswitch Networks and two tiny startups – Tail-F Systems and Affirmed Networks.
Conspicuously missing from the list – at least for now – is Cisco, the world’s largest equipment maker and a firm whose SDN plans remain unclear. The carrier will continue to seek out new vendors, Donovan assures, but the process will be open and inclusive, giving a voice to the smallest of players. Even the network operations management function is set for a root-and-branch overhaul, with more than 1,000 operational support system (OSS) applications facing the chop.
“Our strategy is more than just a network design change”, he says. “It’s a change in how we do business with suppliers and how we manage platforms, systems and software. It changes our very people. We have to take advantage of cultural change at our company.”
And if those words are hard enough to absorb, coming as they do from the mouth of a top-ranking AT&T executive, consider this. AT&T will not seek to develop any overlay technologies while this groundbreaking migration in networks takes place. Instead, the carrier will go straight from legacy systems to the new world.
“We are going to toe-tag the legacy architecture and do our best to rapidly pivot to the new environment. We won’t deploy overlay networks,” says Donovan.
Donovan’s team has already identified what he calls six “beachhead” projects that will test the new architecture and the migration process over the course of 2014. Each of the first six trials involve systems whose working lives will be greatly prolonged by the deployment of SDN controllers, leading to capex savings that will be immediately ploughed back into the Domain 2.0 programme.
By 2015, says the carrier, “a crescendo effect” will lead to the deployment of many more SDN platforms, as more and more capex is freed up.
Not if, but when?
That rather begs an interesting question: is the urgency with which AT&T appears to be driving this project through all about long-term capex and opex savings – Donovan says that the carrier has already seen a “downward bias” on capex costs already – or is it more to do with any perceived early-adopter advantage that AT&T feels it might be able to exploit by attaching such an aggressive deadline to the roll-out?
This question is likely to frame the debate around SDN deployment timetables for months to come, if not years, and it all rather hangs on one simple issue – namely, whether the industry can agree on a single set of uniform standards covering the software language that will power SDN systems.
Some industry players have joined together under the auspices of the Open Networking Foundation (ONF) to promote a uniform language called OpenFlow. But just as many of their rivals argue that SDN should be free from the politics of vested interests and that individual carriers should use one protocol, or even more, to get the best out of SDN, if they so choose.
Crucially, AT&T’s Donovan has so far dodged the issue. The carrier is not a member of the ONF and there has so far been no mention of OpenFlow in any of the launch briefings for Domain 2.0, but the carrier is known to be testing OpenFlow systems.
So herein lies the rub: SDN is not in itself a customer-facing proposition – it concerns itself with control of the guts of the network. Customers are unlikely ever to know or care whether their traffic courses along an SDN-enabled network or not. Many conservative players – including the likes of Vodafone, BT and Orange, might therefore prefer to wait until the industry finally agrees on a comprehensive set of standards, or not, before they finally make a big move.
Others, such as NTT, Pacnet and now AT&T, have grasped the nettle and will worry about complying with standards when, or if, they ever materialise. Given the speed at which some rivals have moved to articulate their own SDN roadmaps since AT&T’s announcement, most notably Verizon and Telefónica, more operators appear to be edging towards the latter than the former.
Further complicating the debate is the extent to which SDN networks can really be monetised. Most carriers accept that the big play in SDN lies in slashing capex and opex costs. The general consensus is that network operators should expect to see opex costs drop by at least 20% following SDN deployment, as computers take on work previously earmarked for skilled engineers.
Yet this might dramatically underplay the real figure. Asian operator Pacnet, for example, has seen opex savings of more than 50% since rolling out an SDN solution last autumn. Then there are the revenue opportunities. Network services are easy to virtualise in an SDN architecture and there will be traffic-optimising opportunities that will help carriers quickly boost their per-Meg rate of return and should help ease the pain of declining voice revenues. New services will be deployable within days rather than months, because the only work to be done will be at the software level, ensuring that SDN-enabled carriers will always be faster to market with new solutions than those running legacy technologies.
Moreover, smaller operators like Pacnet will be able to move at the same speed as Tier-1 players, which could well level the playing field in markets where an incumbent has long enjoyed a competitive advantage over smaller regional players.
But will those revenues be enough to justify junking existing network architectures that still have many years of life left in them? Only time – and possibly John Donovan’s fate – will tell.