Winning the SDN Idol award back in April might not bring extra business to Centec, but it’s a bit of affirmation for the China-based chip startup that’s trying to joust in one of Broadcom’s toughest strongholds — Ethernet switching.
For the moment, the odds are against just about anybody in that fight — just ask Marvell or Vitesse. But Centec thinks it’s finally got a chip worthy of the battle, and CEO James Sun thinks the rise of OpenFlow and software-defined networking (SDN) can give him an opening.
Being based in China has helped Centec build to this point while raising a relatively modest amount for a chip company — $21 million in two rounds, the latest being an $11 million round in 2011.
“What’s impressive is that they’ve done parts that are pretty competitive with what Broadcom has, from a features standpoint, with an incredibly small amount of money,” says Bob Wheeler, an analyst with The Linley Group.
And the Winner Is …
SDN Idol might not have garnered big Nielsen ratings, but it was a big deal to SDNCentral, as it was organized by the site founders Roy Chua and Matt Palmer. It was an unscientific contest held at the Open Networking Summit (ONS) during the after-hours booth crawl, where a panel of experts picked technology finalists and the audience, voting online, selected the winner. Centec won over four better-known names: Big Switch, HP, NTT Communications, and Radware.
To the surprise of no one, Sun credits plain old ballot stuffing for his win. Voting was done via browser, after all, and while an IP address was limited to one vote, there were obvious ways to bend the rules.
But making the final five means Centec at least caught the eyes of the knowledgeable judging panel — and Sun’s presentation won him at least a few fans in the audience.
“One guy came up to me and said he voted six times,” Sun says. “For every browser he had, he could vote once: from his cellphone, his work computer, from home — he even called his wife to vote.”
Clearly, the idea of white-box switching has captured some imaginations — note that SDN Idol happened the night before the Open Compute Forum announced it was adding networking to its charter. And among equipment vendors, there’s a gnawing desire to see an alternative emerge to Broadcom.
Matching Up with Broadcom
Broadcom is the unquestioned leader in Ethernet switch chips, owning, by its own estimate, 90 percent of the market that’s not taken up by ASICs. For growth, Broadcom is trying to pick away at the stubborn use of ASICs inside bit OEMs such as Cisco and Juniper.
Right now, the competition isn’t fierce. Longtime rival Marvell seems to have fallen back in this market. Intel has arrived with the acquisition of Fulcrum, whose long-awaited FM6000 is in the Arista 7150, but it’s going to have to play catch-up. Vitesse has redefined itself as a Carrier Ethernet chipmaker, but its market share is far from fearsome.
Centec has been a non-entity in this fight, having, by its own admission, produced two generations of chips that couldn’t match Broadcom’s features or couldn’t compete on price.
With the third generation, called GreatBelt, Sun thinks Centec has finally hit relevancy. More formally named the CTC5160 family, the GreatBelt devices can outperform Broadcom by about 40 percent, Wheeler estimates, and he thinks the sub-$100 selling price is about half Broadcom’s.
Even so, it’s tricky convincing customers to steer away from Broadcom, a well-known entity. Centec has tried to find an “in” by seeking out particular use cases for its chips, such as cloud applications, but that’s slow work, Sun says.
Some possibilities have arisen with OpenFlow and white-box switching. Centec has been selling chips into ODMs and offers white boxes as well. And it’s a full system reference design, the V350, that won the SDN Idol competition.
Centec spent about four years in lab production, facing a lack of funding that kept its chips two generations behind Broadcom, Sun says. That was followed by one lost year when Centec tried to broaden its product scope while also working on joint developments with big vendors; neither plan worked out, Sun says.
The second-generation product hit the market early in 2011, putting Centec on par with Broadcom, in terms of performance, for the first time, Sun says. Sales didn’t exactly explode, but Centec was able to use this product to get into some customer testing and at least establish a reputation.
That, in turn, set up the GreatBelt product line. Centec claims the chips dissipate 40 percent less power than comparable devices, a claim backed by the use of IBM, with its 45-nm process technology, as a chip foundry.
GreatBelt is designed for 1G Ethernet ports, and there’s a 10G followup due in 2014. It’s that chip that Sun hopes can help Centec work its way into big vendors’ hearts.
“So we are at the point where we have competitive product, we have some level of customer relationships, and we have customer penetration, especially in China,” Sun says.
Where Could Centec Go Next?
The company faces limited options for its future, though. Getting acquired is an obvious exit, but Wheeler wonders if there’s another possibility: becoming a systems company.
“They certainly have all the parts, Wheeler says. “When you’re a chip company, it’s pretty tough to compete when you’ve got this single product against Broadcom with its portfolio. I’ve seen this story before, and a lot of the time, these little companies find out it’s a lot easier to generate revenue by selling boxes.”
“They should just become a systems company,” he says. Centec is already providing some systems-level products to Tier 2 players, including “one customer in Korea that I think is fairly far along,” Wheeler says.
The one partnership Centec has announced is with Vello. Interestingly, Wheeler says Centec doesn’t seem to have made made progress selling to Huawei or ZTE, two obvious candidates.
For now, Sun is banking on SDN and particularly on OpenFlow, having jumped onto that bandwagon early. Companies such as VMware and Cisco, and initiatives such as the OpenDaylight Project, are emphasizing that SDN doesn’t have to use OpenFlow, but Sun doesn’t think that paints Centec into a corner.
“I kind of agree the overlay model is getting better market adoption, but having said that, OpenFlow still plays a role in that overlay model,” Sun says.