Blog Post

Pica8: A startup taking advantage of network commoditization

Stay on Top of Enterprise Technology Trends

Get updates impacting your industry from our GigaOm Research Community
Join the Community!

Networking is having its heyday in Silicon Valley as startups pushing programmable networks raise millions, and venture firms are on the lookout for the next big thing in IT infrastructure. There are three big trends coalescing in the space that are driving this excitement, and Pica8 hopes to take advantage of them all.

One is the rise of merchant silicon that can deliver the performance of custom chips made by the likes of Cisco and Juniper (s jnpr) but at a lower cost. Another is the rise of software-defined networking, which separates the physical network infrastructure from management of that network. Finally, networks are getting bigger, and doing so faster, which is creating pressure on systems adminsitrators to find a way to scale their networks faster and buy the gear more cheaply.

Pica8, a two-year-old startup that spun out of server maker Quanta last month, hopes to become the anti-Cisco (s csco) by pushing networking gear made on merchant silicon available from Broadcom, Marvell and Fulcrum. The promise of cheaper boxes is enough for some folks, but James Liao, the CEO of Pica8, has a more disruptive plan — to build software for the white-box networking gear and then open that up to the industry as a whole. It’s taking one step further the vision of Arista, which began in 2008 and puts its own software on top of boxes it builds using merchant silicon. And in the process, Pica8 will shave away at industry margins.

Liao wouldn’t disclose the price of the Pica8 boxes, but sources say they will run from 20 percent to 60 percent less than gear from Cisco, Force10 (s dell) or others. Already, Pica8 has customers, including NTT communications and Baidu, as well as a strategic partner in Quanta, which is making the Pica8 boxes. Pica8 will also work with another Taiwanese or Chinese white-box maker to deliver its software to another vendor.

Liao would like to work with more, but with 15 engineers and himself, he’s strapped for resources. To help, he’s hoping to complete a $5 million round of funding in a few months that will help the company expand.

The new networking trend

Networking has been somewhat of a black box for businesses for years. The boxes contained proprietary chips known as ASICs that were able to handle the speeds required to send packets around a network, and used proprietary software that required specialized skills to program. All this specialization made networking both pricey (for customers) and profitable (for providers). But about two years ago, the chips from providers like Broadcom and Fulcrum, which Intel (s intc) bought, began to match that quality and performance of the specialized ASICs made by Juniper and Cisco.

And with those chips came an opening for a change in the network. Arista picked up on it and began selling boxes built on the commodity silicon, while Google (s goog) built its own switches using merchant silicon as well. With these cracks in their chip business, it became possible to build boxes that didn’t require a multi-million-dollar chip investment inside. From there, it becomes a question of software.

Commodity hardware opened up the chance for a vendor like Arista to differentiate with its software. But with OpenFlow controllers and companies like Pica8 pressuring the proprietary software running on networking gear, many companies are seeing an opportunity to get in on the ground floor of a revolution. Pica8 hopes to be one of those startups, and with backing from Quanta, which itself is disrupting the server business, it’s one to watch.

5 Responses to “Pica8: A startup taking advantage of network commoditization”

  1. Linux Guy

    I’ve just read everything I can about Pica8 and I’m excited. As a linux sysadmin with over a decade of experience (and an ex-googler), I can’t wait to see the open-source network software quickly catch up to the incumbent network providers. And with inexpensive gear available to developers, I expect Xorplus to because feature rich, stable and reliable real quick.
    Expect LOTS of dismissals from the cisco’s, arista’s, force10’s of the world. pica8/xorplus threatens their entire business model. This commoditization of network gear has been a long time coming, especially with things like Hadoop driving up network cost to like 20% of the colo build.
    Can’t wait for it to drop back down.
    Pica8 has it right. I’ve held all the fancy switches in my hand. They Used to be built differently, but today its all outsourced to the lowest bidder, and pica8 is getting boxes off the same ODM assembly lines.

  2. Daniel Golding

    Every few years, there is a startup who tries to do this, going back at least ten years. There is always a well-baked piece of open source software that can do wire-speed routing at current peak speeds – except when it doesn’t. Low speed routing is easy and cheap already.

    Ask yourself why HP hasn’t eaten Cisco’s lunch in the networking area? Its because what Cisco and Juniper charge the big bucks for is HARD.

    Usually startups like this do well if they concentrate on a very specific area of functionality and make it both wirerate and inexpensive. The exit is a strategic purchase. However, generalized open source network-in-a-box is a nonstarter – especially with 15 coders.

  3. Lorenz Redlefsen

    I would not put products into the “commodity hardware” category simply because they use merchant silicon. System design is a hugely important, but often under-appreciated, quality.

    A good analogy is the PC business, where everybody starts off with the same x86 chip. I think we’d all agree that an Apple Macintosh is not “commodity hardware”, for example — there is a huge difference between a Mac and other x86-based PCs in terms of system hardware, and that’s before you’ve even started to talk about the software differences.

    Unfortunately, it’s very difficult to show this difference in a side-by-side comparison on paper, where most of the focus is on properties of the forwarding chip, e.g., table size, port count, etc.

    One place to look is whether the components that are most likely to fail (power supplies and fans) are redundant and/or hot-swappable. How long can the system continue to run when a single one of these components fails?

    Another thing is to just hold the hardware in your hand, just like you have to test-drive a car to actually appreciate the differences.

    Full disclosure: I work for Arista, so I know what to look for in a well-designed system, but I am stating my personal opinion — I am not speaking for my employer. :)