Doug Gourlay, Arista Networks: Programming the future of SDN
There is no future in a network that cannot be programmed, believes Doug Gourlay, VP of technology marketing and systems engineering at startup vendor Arista Networks.
It was as an eight-year-old boy growing up in small-town Nebraska that Doug Gourlay discovered the joy of tinkering with computers. A love of writing elegant software programs was born, and soon he was creating a bulletin board system to help different computers talk to each other more efficiently.
“Programming seems to have been a theme of my life,” he muses. Now VP of technology marketing and systems engineering with equipment vendor Arista Networks, Gourlay has become a passionate evangelist for software-defined networking.
Arista is making its mark on the SDN market in the classic fashion of a disruptive startup, unencumbered by the dead weight of a legacy portfolio. Its senior management team is made up of alumni from a spectrum of major networking brands, but its thinking is pure blue sky.
The company’s innovative Extensible Operating System (EOS) offers single-image consistency across hardware platforms, enabling in-service upgrades and application extensibility, and crucially, says the vendor, changing the price performance model for data centres and their network operator partners.
Defining the SDN universe
As passionate as Gourlay is, he rejects the idea of a purist SDN universe in which all networks are created to one centralised ideal.
“First we need to build a clear picture of what we mean by SDN,” he explains. “Some people have got a more fixed idea of what it is than others. You’ll hear some say ‘OpenFlow is SDN – and nothing else’. I think that kind of attitude is fading now, and people are seeing SDN as any kind of progress along a path where a network infrastructure is programmed to create business value.”
Carriers, he believes, should be taking as broad and open an approach to their SDN transition plan as possible, and would do well for an example to look at how some of the big data centre players have taken it on board – the Googles and the Amazons.
Openness has been their key to unlocking business value from SDN. Network operators, claims Gourlay, have much to gain from successfully negotiating programmable networks.
“If you get it right, then bringing a new customer online takes seconds, since all your network provisioning works the same way,” he says. “It’s all automated, which is a significant change to when everything by necessity had to involve a human sitting at a keyboard. The aim needs to be a non-proprietary network using non-proprietary programming protocols.”
But Gourlay acknowledges that the vendor world is still populated by apologists for the monolithic, proprietary approach.
“If a single vendor is trying to say they can solve all carrier problems with their solution, then ultimately that vendor probably hasn’t got the carrier’s best interests in mind,” he insists.
“You’ll hear all sorts of claims. I heard someone from a large old-school vendor say ‘We do SDN management’ the other day. It was the kind of vendor that’s got technology that’s 10 years out of date and is struggling to stay relevant. Since when isn’t all network management software-defined?”
The battle of the vendors
Gourlay believes all the major vendors are coming at SDN in their different ways – some more credibly than others.
“You’ve got Juniper making acquisitions, and Alcatel-Lucent with its pet in-house project,” he points out. “Saying ‘We do SDN’ is probably the best way of getting a meeting with a senior network operator executive right now.”
He believes the history of large networks shows that openness ultimately prevails, and maintains that Cisco won the routing wars of the 1990s because no other vendor offered support for as many protocols.
“Customers looked and saw that any decision they made would be supported,” he says. “The vendor with the most interoperable and open solution eliminated barriers for their customer and simplified their choices. The monolithic approach of SGI and Sun was suddenly not winning.”
The survival of the proprietary monolith may be, says Gourlay, because “lots of customers aren’t that smart”. Any buyer of network solutions should always be asking who offers the best design, he insists.
“Some customers will lock themselves into one vendor because it makes their job easier that way,” he sighs. “But IT’s job is not to make IT easier, it’s to support the business.”
He believes while some equipment makers are battling to accommodate an SDN future, others are simply running scared.
“Lots of vendors are unaligned to what SDN wants to do,” he says. “If you’re a firewall company then it’s not exactly exciting to hear all about how firewalls can be virtualised. It’s scaring them, quite rightly.”
Standing out in the SDN space
Arista is consciously cut from a different cloth from most other vendors in the SDN space. Its switches support several different architectures, the broadest portfolio of any vendor, it claims.
“This gives the customer a lot of choice,” says Gourlay. “We address 95% of the North American data centre market with one common design. We support data centres, cloud companies and Web 2.0 companies with a consistent architecture, with just different software and provisioning to define how each operates.”
Arista’s crown jewel is its EOS operating system, designed to be as programmable and unmonolithic as possible.
“It gives the customer the ability to do very interesting things,” explains Gourlay. “Conventionally when upgrading, you type ‘reload’ and the switch turns itself off, forcing a hard reload. But is that what the customer really wants? A programmable design takes away that problem and gives you a smarter system upgrade with no disruption or outage. For carriers this is incredibly important. An outage can get you on the front page of the Wall Street Journal and we don’t want to be the vendor who does that to you.”
The vendor set out to be the most constraint-free network vendor possible, he claims.
“Our goal is an upgrade with no extra software deployments and no hardware changes. Changing the Command Line Interface is anathema to all customers,” he says.
Gourlay is reluctant to make wild forecasts about where the SDN market is headed. He is not by his own admission a gifted market analyst. He believes the market will find its own level at a pace essentially determined by the paying customer.
“Three years ago, I heard pundits saying OpenFlow would take over and become the only way to do networks,” he says.
“The reality has been very different. People are building networks traditionally and looking to SDN to make what they do more efficient and automate what they can. I don’t foresee much of a standalone SDN market. If you predict that SDN will be worth $30 billion, then you’re mostly just recounting existing network spend. As for the technology, I think we will see most networks move to a framework that some will call SDN, on the basis that it involves programming the network in some way.”
Whether the future of SDN is interpreted as simple or complex depends on which way you look at it, he feels.
“Network architectures will be simpler and flatter, designed to be programmable,” he expects. “Jobs that took weeks will take minutes. But there will be no one evolutionary path for all participants in the network industry. Any vendor setting out to go galactic will have their work cut out. It worked with IP, but it won’t be that way with SDN.”
So what does someone who programmed computers as an eight-year-old do for fun? Get out and have fun, as it turns out.
“Every year I like to set myself a new challenge,” says Gourlay. “I try to learn a new life skill that will help to make my retirement fun. One year I learned to fly, another to sail. One year I gave to improving my golf game. This year I’ve been biking. I wanted to find some therapy for my leg after I shattered it skiing and running wasn’t the answer.”
He admits to spending much of his working life “running to keep up with the scary-smart young technology guys in the company”.
But such is his passion for technology and for the crafting of software that this probably counts as fun too.