This is the latest in a brief series of articles about white-box switching. Earlier installments profiled Cumulus and Pica8 and covered a new open-source switch from Centec, and the white-box business model.
At least two startups are working on Ethernet switch chips, which might indicate how seriously the market (and venture capital) believes in bare-metal switching.
Xpliant and Barefoot Networks are both founded by chip-industry veterans. Both are still keeping their plans quiet, but Xpliant has said it’s designing switch chips, and an industry source says Barefoot is doing the same.
Semiconductors haven’t been a popular Silicon Valley investment for years. Moore’s Law has let chipmakers pack more circuitry per square millimeter, but as feature sizes shrink, the chips get more expensive to manufacture. Chip development is also a slow process, especially compared to the quick turnaround time of mobile-phone apps.
That’s why it’s remarkable to see a chip sector spawn multiple startups. Moreover, the experience of Centec suggests that with help from inexpensive labor in China, a switch chip can be built with an investment of just $21 million (so far). That’s more expensive than an apps startup but not the $100 million you sometimes hear quoted for high-end-chip development.
Xpliant was founded by former Marvell employees including Tsahi Daniel and Sachin Gandhi, Xpliant’s CTO and COO, respectively. Founded around 2011, the company has been participating in the Open Networking Foundation (ONF), and Daniel serves on the recently formed Chipmakers’ Advisory Board. Even though it’s in stealth mode, Xpliant hasn’t exactly been invisible.
Barefoot, meanwhile, has its roots in Texas Instruments. Martin Izzard, who’s reportedly running Barefoot, spent more than 20 years at TI before leaving in mid-2013, according to his LinkedIn profile. Assuming Barefoot is this company, it ran an online contest to design its logo. At least they’re having fun at this.
Why Broadcom Will Keep the Lead
Having multiple chips on the market is nice, but that doesn’t mean they’ll be easily interchangeable. Each company has its own software development kit (SDK) that a systems maker has to learn if it’s going to tie any features into the chip.
On top of that, some of the OEMs (the non-white-box vendors) worked individually with Broadcom on their systems designs, and they even gave some input into the chip design for Trident II, Broadcom’s newest high-end switch. That collaboration makes it harder to move to a competing chip, but it can also give the systems vendor a competitive edge.
“You can expose features quicker, you can design better, and you can work around issues that are in the SDK. You can work with the silicon vendor on how the SDK works — or how it doesn’t, sometimes,” says Todd Acree, director of product line management at Extreme Networks. “A lot of times we rewrite part of the SDK because it didn’t work the way we want it to. Maybe that’s an area where people will provide off-the-shelf support eventually. I don’t know.”
Broadcom’s dominance in Ethernet switching lately means some customers have been comfortable with pegging big plans to its chips.
“We have at least one customer that’s looking at a particular type of software-defined networking — hyperscale fabrics, firewalls, load balancing, and other security types of products. The architectures are not just a simple spine/leaf. There’s a lot of diversity in how the architectures are being built out,” says Rochan Sankar, director of product marketing at Broadcom.
How Broadcom Might Lose the Lead
Of course, Trident II must have its flaws, and one is obvious: The chip was late in ramping to volume production, according to multiple customers.
Broadcom isn’t elaborating on that. “The switch is now in volume production, and we feel sets the stage for growth in the next year. I don’t really have more to say beyond that,” Sankar says.
The delay revealed the pitfalls of relying on external chip development. Yes, ASICs can miss schedules too, as is rumored to have happened to Insieme. But one reason to use an off-the-shelf chip is to save time as well as money, and you don’t get a “time refund” if the chip is late.
Intel and Mellanox both want to make a dent in Broadcom’s switch-chip dominance, and the delay might have given them some time to catch up and to win more attention from some customers. Trident II has won the current generation of switches, but it’s no stretch to believe that other big companies, already eyeing the networking space, are trying to exploit Broadcom’s misstep.