Intel Has Its Eyes Fixed on Networking
You know Intel’s long history of supplying the main chips that run personal computers and servers. And you probably know that it supplies some chips for mobile phones and tablets, though not quite as many.
But did you know that Intel is also getting into the business of supplying chips for networking gear? The way the chip giant sees it, there’s a $16 billion business of supplying processors for gear that moves data, whether it’s in a corporate data center or on a telco carrier’s network — or anywhere in between. And it reckons that it already has about a five percent share of that business.
The way it usually works is that a networking vendor like Cisco Systems builds its own custom chips, called ASICs, that go inside its routers and switches. But now there’s a shift brewing toward software-defined networks — networks where the hardware is commodity and the setup is all done in software. That’s one place where Intel may have an advantage, the VP and general manager of its networking group, Rose Schooler, told me in a recent interview.
Last month, Intel debuted a new chip aimed at networking, code-named Highland Forest. It’s a version of its Xeon chip for servers that is six times faster than its predecessor.
My first question was about that $16 billion business Intel says it sees, and how it defines that market.
Re/code: Rose, let’s start at the top. You claim to have a pretty significant market opportunity here. But I’m curious as to how Intel defines the networking business you intend to play in.
Schooler: When we say networking, it’s a pretty broad definition. We mean anything that moves data. That’s in enterprise IT, in the cloud, in wireline, wireless and cable networks. It could be a top-of-rack switch in a data center, it could be a network appliance like a firewall or intrusion detection system. It could be a base station, or a gateway within an operator’s network. Our targets include all layers and all industries.
So it would seem to be kind of a tough business for Intel to crack if, for no reason other than there’s a lot of well-established players in it (including Cisco and Juniper) and they tend to design their own specialized silicon.
We look at the needs the end users have and we try to address them. So we started looking at the telecom companies in particular, and what we found is that as they built out their networks to support the breadth and volume of data that’s being driven by the introduction of the iPhone and other smart phones …
They’re under pressure to expand their networks and want to simplify somehow?
Yes, but it’s worse than that. As they were doing their long-term forecasts, they started to notice that the cost of implementing those networks would eventually exceed the revenue they would generate. And when they began to look at ways to solve it, it boiled down to two problems: How do you save money on capital expenditures and operating expenses? And then, how do you create new streams of revenue with new services? So when we started to craft our networking business strategy, those were the problems we aimed to solve.
So clearly you want to get in on this business as this transition happens. Where are your entry points?
There are really four. The first is the move to general-purpose computing by the telcos. Since about 2000, we’ve been investing in things like carrier-grade Linux; at about that time, there was a migration by the telcos away from proprietary systems to more industry-standard servers, often with Intel processors. When they did that, they saved a lot of money, and we’ve seen a lot of proof-points that bear it out.
The second is processing in the control plane. Once we got multi-core chips, we could get into the low-power realm, and we could do it in a way that was affordable. The control plane is really where all the signaling happens in a wireless network. That gives us an opportunity to win there.
We also have some opportunities in the data plane, handling the packets themselves. We used to develop and still sell network processors, but we’ve taken what we learned and put them with standard silicon, and when you combine it with software you can get some really great performance.
Finally we’re using our graphics capabilities for things like media transcoding. The telcos have been coming to us and asking for this.
So where does software-defined networking fit in all that?
SDN is an approach to get more flexibility and scalability, and it shortens the amount of time it takes to set up new services. And what we’re learning from our own internal pilots is that where it used to take weeks, if not months, to deploy a new network capability, now — from our own experience — we’re doing it in minutes.