Intel missed the boat on mobile, and it may be facing a fight from ARM-based chips in the data center, but the chip giant isn’t going to sit back and lick its wounds. It is actively reaching out to the burgeoning maker and connected device crowd with development boards, it launched a new chip designed for the internet of things and it purchased an API management company. It is also going after a less sexy but still large market that seems ripe for some x86 chips — telecommunications.
Intel wants its chips inside the switches, routers and base stations that are the workhorses of the telecommunications network, even if it means making some sacrifices. To meet telecom’s requirements it will have to partner with another chip firm to get the right kind of processing power for base stations and change its focus from integrating everything into one chipset into a more modular system-on-a-chip model. In a conversation with Rose Schooler, the GM of Intel’s Communications Infrastructure Division, we discussed the chip firms plans for what it sees as a $16 billion market opportunity, why the telco market is ripe for disruption and what Intel has learned from its previous (and abandoned) forays into this market.
The emergence of Telco 2.0
But first, let’s talk about why this is even an opportunity for Intel. Today, the telco market has a big problem in that people love its products (especially wireless data) but the cost of meeting that demand is too high. One reason for such high costs is the type of equipment telcos purchase. The concept of “five-nines reliability” (available up to 99.999 percent of the time) and telco-grade gear have created a booming business for expensive boxes that handle much of the telecommunications industry’s needs. The dominant chips in this market are PowerPC or proprietary network processing chips from networking gear vendors.
But much like the server industry in the 1990s, when Intel saw an opportunity to move its PC chips upmarket, the telco world looks ready for a cheaper, good-enough chip as long as it can meet certain minimum specs. Already telcos are trying to consolidate their hardware (it’s not uncommon to see telecommunications providers have their internal IT, their IP video equipment and their cloud services all on three different platforms) on a common infrastructure to offer them more agility.
As this transition happens, Intel (as well as ARM) is trying to offer an architecture onto which telcos can consolidate. When you add in newer trends such as software-defined networking, the telecommunications industry is at an inflection point that all kinds of gear makers are trying to adapt to. Intel wants to be in on that, and has for a long time.
“This isn’t a strategy that started when SDN started to get attention,” Schooler said. “We’ve been on this journey for a decade. We’ve been looking at the evolution for the IA [Intel architecture] franchise within communications and redefined networking to boil it down into all of the infrastructure and workloads that support a network function.”
Understanding the workloads
Intel believes there are four different workloads that its chips can address. It has better hopes of success in some of these while Schooler admits it would need a partner in others. The segments are:
- Application processing: These are servers running traditional telco workloads like billing. This is an area where Intel already dominates in other enterprises so switching telcos to a commodity server architecture isn’t far-fetched. Investments include spending on carrier-grade Linux.
- Data plane processing: The data plane is how a router decides where to send a packet. This generally involves looking up information in a routing table that contains the network topology. Specialized processors from Broadcom(s brcm) and Cisco(s csco) and Juniper(s jnpr) handle the bulk of this work inside telcos today. And in the giant core elements that still will be the case, said Schooler, who noted that Intel isn’t going after the routers that handle the largest amounts of traffic. Frankly, it can’t: Its chips just aren’t fast enough. It’s current chip for this market, The Crystal Forrest chip handles 160 million packets per second.
- Control plane processing: This is the physical work of moving bits around the network. For a long time, control plane and the data plane processing were combined in a single box, but technologies like Open Flow are leading to a separation between the two.
- Signal processing: This area encompasses the radio access network or RAN, and this is where Intel is the weakest. Processing cellular signals requires specialized digital signal processors as well as chips that are able to convert analog signals to digital and sometimes back to analog. Intel clearly has an eye on this market, as my colleague Kevin Fitchard covered last year, but Schooler admitted that it would have to find a partner.
Handicapping Intel’s odds
The challenge for Intel will not be proving that the market needs a more agile, lower-cost product, but that Intel’s architecture is the right one for the job across all four workloads. On the application side, it’s in a fine position, but on the control and data plane side it will need to develop an x86 chip that can offer the throughput of a network processing chip as well as features that telcos care about. This will require a new chip, and Schooler didn’t want to discuss this chip.
As for the signal processing, Intel has a deal with China Mobile to try to build out a Cloud-RAN, but China Mobile is also testing gear from Alcatel Lucent, an established vendor that has a partnership on the chip side with Freescale. That partnership seems to be going strong.
Schooler noted that in signal processing, Intel will likely have to rely on some-fixed function IP that comes from outside of Intel, which has me wondering where Intel will find the special-purpose compute engines to handle the tasks one taken on by DSPs.
In the core of the network, she’s more optimistic, pointing out that Intel has done a lot of work to upgrade its CPU architecture to rival the packet-moving prowess of a network processor. Schooler seemed to understand, however, that Intel will have to rely on proprietary networking chips (ASICs) as well as its acquisition of Fulcrum, a networking chip provider that Intel purchased in 2011.
“In the transport there will be a network processor and ASICs and the CPU will not take over all cases of throughput.” She pointed out that Intel has improved its throughput on its chips by 10x over previous generations, but she also knows that the fatter switches and routers that handle lots of traffic aren’t candidates for the Intel architecture today.
I’ve long said that you shouldn’t bet against Intel. This is a company that may miss a market, but it has the capital, the research and the market share to force its way into a business and make a go of it. Fundamentally, Intel bets on cheaper general purpose processing, which is how the market has moved over the last couple of decades.
It somehow missed the importance of lower power architectures, but is now doubling down on its commodity computing bet for the telecommunication market. We’ll see if the telcos are ready for that.