The project, in a presentation by Frank Frankovsy at Interop, said it was taking a slightly different tack with its design, deciding to get input from others before actually making and releasing the hardware to the community. However, just because the hardware isn’t designed yet, Facebook isn’t going to twiddle its thumbs for a traditional multi-year design cycle. Frankovsky told me in an interview that he expects the hardware to b out in 9 to 12 months.

“We have built these islands of openness in the data center but the last element, and the one that was connecting the compute and storage, was the network,” said Frankovsky. “And there is a lot of pent-up passion out there for breaking open this appliance model.”

Networking is the last bastion of proprietary profits

Prepare to be disaggregated, switch!

For those who don’t dwell in data centers, the top-of-rack switch is the networking gear that sits on the top of a rack of servers directing traffic between those boxes and between the other racks in the data center. While the networking world is all aflutter over the promise of OpenFlow and software-defined networking, very little real progress has been made in building switches for the webscale data center.

Google, a few years back, had famously issued a request for a new type of switch that would fit its very specific scaled-out needs and no one responded. Now the search giant makes its own hardware. But soon after that, Andy Bechtolsheim saw the need for Google-like speeds and scale and started Arista, a switch company that has dominated in the webscale, financial and high-performance switching space. Meanwhile, at the lower end, Cisco’s cheaper Nexus line of switches have done really well.

Facebook’s Najam Ahmad.

Yet, these options aren’t palatable for Frankovsky or Najam Ahmad of Facebook (Ahmad will be at our Structure conference in June discussing more about Facebook’s networking strategy). On the existing product side, Frankovsky is frustrated by hardware that doesn’t play nicely at scale. He specifically mentioned that the side venting of heat on switches means he can’t place them right next to another switch. Ahmad, who is in charge of the social-networking giant’s network, is concerned about getting out of the proprietary OS model.

“We want it to be OS-agnostic so we can use one from our existing provider or build our own,” he said. He added that he’d prefer an open Linux-based implementation. These proprietary OSes — Cisco has IOS, Juniper has Junos and Arista has EOS — are one of the reasons that companies are locked into one networking gear provider. They are also stuck using proprietary code to make changes.

Who will be the Red Hat of the networking OS?

Networking cables along the ceiling at Facebook HQ.

If you are chock full of technically savvy people, losing the agility that comes from writing your own code as well as paying higher prices for the proprietary hardware and software combination is probably maddening. Hence Facebook’s interest in the open source OS. Of course, building out the underlying hardware is only the first step, the next will be supporting an OS that runs on top of that system.

While Facebook might build its own OS, not every company will want to do that, and Facebook may not open source its own networking OS if it ever makes one. That leaves a market opportunity. Perhaps a firm like Arista might move in here with an open source version of EOS, although given that Arista uses merchant silicon in its boxes, putting up an open-source version of its software would eat into its margins.

This is neither Open Flow nor SDN

But let’s go back to the box. Facebook is working with Broadcom, Intel, The Open Daylight Foundation, the Open Networking Foundation and Big Switch as some of its collaborators on this project. The box itself might run x86 hardware or a proprietary ASIC, according to Frankovsky. As for the protocols, Open Compute is going to see what the other collaborators want.

Software-defined networking

But for those wondering about Open Flow support, it’s likely. Frankovsky said that the Open Networking Foundation asked Facebook to get involved via the Open Compute Project with making open networking hardware. While Frankovsky and Ahmad didn’t cop to it, I know there has been frustration in many areas of the webscale and networking world that the promise of commodity hardware that Open Flow could offer has not really hit the market in a way that offers the most flexibility for data center operators.

Frankovsky said that the ONF approached Open Compute (Facebook is a founding member of both organizations) in part because it believed it could move quickly on this. And it will. But it’s worth noting that this announcement is about an open source top-of-rack switch, not a controller and not some type of software-defined networking play.

Other companies may take this box and perhaps an open source OS if one is developed, and then layer on some type of controller software to make a software-defined network, but this is just a box.

That being said, this is a box that could seriously disrupt the existing players in networking, from giants like Cisco and Dell all the way to smaller startups like NoviFlow or even Pica8. Much like Facebook is changing the server market with Open Compute, we’ll see if it can tweak the model and do the same in networking.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• Cloud computing infrastructure: 2012 and beyond
• OpenFlow and beyond: future opportunities in networking
• The promise of SDNs in the enterprise

Software-defined networking doesn’t require OpenFlow. And it will still make a network programmable and responsive in ways that both scaled-out web-services providers and enterprise customers dealing with virtualization will appreciate. But it won’t necessarily affect the underlying networking hardware in the same way OpenFlow can. However, OpenFlow — the protocol that aims to separate the intelligence required to route a packet from the act of moving a packet– can commoditize the switches and routers. And it will have a big impact on the networking vendors such as Cisco, Arista and Juniper.

SDN is good for many things, but not everything.

Nick McKeown, one of the fathers of OpenFlow onstage at Structure.

But it’s early: the promise of real, commodity-based networking gear will not dissipate. Stu Bailey, the CTO of Infoblox told me last week as his company was launching new software-based networking products that the emphasis on specialized chips and networking hardware is doomed. Both cloud and the enterprise networks are becoming increasingly complicated; not just because of virtualization and scale, but also because of the increasing number of devices at the edge. Smartphones are pressuring networks, but adding sensors and other connected devices as integral business equipment, will require some serious thinking about networking hardware and cost.

“How quickly will a large healthcare org realize that the network is not the collection of routers and switches, but is instead these things connected to the network?” asked Bailey. “And how they interact fundamentally and how they are secured is responsible for the business. With that awareness they need SDN economics and OpenFlow will hasten that.” Hence InfoBlox’s decision to focus on software — software that is OpenFlow compatible.

This revolution won’t come immediately, but it will come.

Others, such as Jim Theodoras, director of technical marketing at ADVA, an optical provider to data centers, have expressed similar thoughts. Theodora has also expressed frustration that so far the message of router and hardware commoditization has been ignored. But for the most part, the existing products on the market and their associated marketing and product managers have been good at loudly shouting down the camp of people who see OpenFlow as a legitimate threat to the hardware makers. And it’s not just marketing. There are many obstacles to deploying an OpenFlow-based networking infrastructure.

(L to R) Lane Patterson, CTO, Equinix; Kenneth Duda, Founder, CTO and SVP, Software Engineering, Arista Networks
(c)2012 Pinar Ozger pinar@pinarozger.com

One obstacle, other than just getting the gear to work, is that the current chips sold by merchant silicon providers aren’t ready to support the most recent and faster versions of OpenFlow. NoviFlow, a startup in Montreal, that just announced an OpenFlow compatible switch that processes information at 100 gigabits per second per second — a significant amount of capacity when we’re talking about the data center market — went with specialty networking processors. It couldn’t build such a high capacity OpenFlow switch with existing chips, so it used network processors from EZChip. It’s still using a specialty chip to make its boxes, much like Cisco and Juniper do.

OpenFlow creates a common networking platform, not lock-in

NoviFlow’s 100 Gbps switch.

Marc LeClerc, NoviFlow’s VP of strategy and marketing is anticipating that day, explaining that NoviFlow has a high-end switch, but that it expects customers to also purchase lower-end gear that they will expect will work with the NoviFlow products. And when that day comes, the question is what kind of shakeout we’ll see in the networking world.

“When the world went over from mainframes to client-server it was a huge shift and plenty of companies that used to play in that market like Wang and DEC are no longer around anymore,” Bailey said.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• Cloud computing infrastructure: 2012 and beyond
• The promise of SDNs in the enterprise
• Infrastructure Q1: IaaS Comes Down to Earth; Big Data Takes Flight

After two years of trying to figure out what OpenFlow was and how it might change networking for the better, 2012 was when the promise of OpenFlow morphed into software-defined networking, which was later co-opted into network virtualization. Things are still heating up, but let’s look at where we’ve been.

Guido Appenzeller (left) and Kyle Forster of Big Switch

This was the concept of SDN. Under that title, the physical hardware of a network was abstracted from the virtual machines and applications running on the network. Sometimes OpenFlow might have a hand in this and sometimes it won’t.

Instead of commodifying the router, these companies added a layer of software between the networking gear and the application, generally known as the controller. Some of these were open source, some provided APIs for the accessing the underlying networking gear and some did not. The point in SDN was that once you had this virtualization, it became possible to link your network to your application. Depending on whose controller you used, this process was easier or more difficult.

Dante Malagrino, Co-Founder and CEO, Embrane; Jonathan Heiliger, General Partner, North Bridge Venture Partners; Martin Casado, Co-Founder and CTO, Niciria
(c)2012 Pinar Ozger pinar@pinarozger.com

With that framework in mind, here are the notable deals in the networking world in 2012.

Oracle buys Xsigo: This deal, which was announced in July, wasn’t really an SDN deal. Xsigo has really tried to grab ahold of the SDN banner with its marketing, but it was about the virtualization at the hardware and port layer. Xsigo’s hardware (plus fabric) make it possible to plug storage and networking cables into a Xsigo box and then allocate those physical resources without a network engineer having to get involved. Oracle buying Xsigo fits with its proprietary hardware plans and less with any sort of Oracle SDN play.

Nicira’s CTO Martin Casado

Brocade buys Vyatta: Vyatta is another company that tried to give itself an SDN facelift. In this case, Brocade, the company better known for making switching hardware took the bait (or was simply merciful), and said it would buy Vyatta while the company was making the rounds trying to raise another round of venture capital. Brocade can use Vyatta’s networking software to move up the stack and possibly provide a more modular approach to SDN that allows customers to piece together Broacde’s hardware and software if they so choose (but they won’t have to choose both as they might in a Cisco or Arista decision).

Cisco to buy Cariden for $141 million: In November, Cisco purchased a network mapping and optimization company that has recently adopted some SDN features (and marketing). For Cisco, buying Cariden made sense because it delivered the type of controlled network virtualization ecosystem that posed no danger to Cisco’s router business, but allowed Cisco to sell SDN services and applications to its service provider customers. Meanwhile, Cisco this year also announced its own controller layer and an API to let applications talk to it (but not the underlying Cisco gear.)

Juniper to buy Contrail for $176 million: Finally, last week Juniper said it is buying Contrail, a startup in the SDN space that had only launched a day or two prior to the acquisition announcement. The deal came just a few months after Contrail had received $10 million in funding from Khosla Ventures, with Juniper participating as a strategic investor. Contrail’s technology centers around rethinking where the controller would sit in a virtualized network and how those pieces work together. It had a great team and good-sounding tech, but few real customers.

The year isn’t over and it’s possible one of the myriad networking startups funded this year, such as Big Switch, Pluribus Networks, Pica8, Plexxi and/or Cyan might end up getting bought before we close the books on 2012. But even if that doesn’t happen, I’m comfortable saying this is the clearly the year SDN has sold out and networking has really broken out.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• The promise of SDNs in the enterprise
• Cloud computing infrastructure: 2012 and beyond
• The promise of software-defined networking

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• Cloud computing infrastructure: 2012 and beyond
• Infrastructure Q2: Big data and PaaS gain more momentum
• Infrastructure Q1: IaaS Comes Down to Earth; Big Data Takes Flight

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• Infrastructure Q1: IaaS Comes Down to Earth; Big Data Takes Flight
• Infrastructure Q2: Big data and PaaS gain more momentum
• Infrastructure Overview, Q2 2010

Surprisingly, the audience willing to spend big bucks to reduce latency by seconds –or milliseconds–might be larger than you could imagine. Solarflare’s background is in the world of high-frequency trading where milliseconds can cost millions, and large financial players were willing to shell out for Solarflare’s network cards. HPC networks and superfast academic networks are also customers.

On Tuesday the company added a new twist to its offering by adding a specially designed networking chip and middleware to its adapter cards that allows a customer to connect an application directly to the network. So for applications that need it, Solarflare’s new cards can allow a customer to process data coming in from the network in real-time. This is a big deal for its existing markets as well as for any company looking to process large amounts of data in real time.

The product called the ApplicationOnload Engine has a silly name, but its a powerful concept. It combines a specialty chip called an FPGA, with Solarflare’s adapter and with its middleware (see diagram) on a single card that can be slotted into servers. The whole package is designed to make programming the chip easier without sacrificing speed.

FPGAs were once common in the networking space, and vendors spent millions designing them to eke out performance gains for their gear. But while hardware offers faster performance, it’s a pain to program, which meant that FPGAs weren’t user-friendly or flexible. As general network services became more attractive than speed, network vendors tended to neglect the smaller market whose need for speed trumped general purpose networking.

Much like Arista, which serves that high-end market with its own switches and software, Solarflare is hoping to pick up business where the major players have left a hole. Russell Stern, the CEO of Solarflare, says Solarflare’s financial customers are ready to trust their networking applications to Solarflare after relying on the company’s cards for the last few years, which prompted the move into this next level of service.

Stern also sees a potential market in big data processing and even new use cases such as enabling social networks to conduct auctions to deliver real-time advertising at the moment when a user refreshes a page. Much like Fusion-io — which had conducted a successful initial public offering based on its premise that a large class of companies would pay for a separate adapter card that helped boost the speed of solid state drives and added intelligence – Solarflare is catering to a once-niche market that is growing and underserved by the larger vendors.

As we demand faster page load times, faster networks and faster transactions Solarflare is a natural beneficiary. Solarflare’s networking cards aren’t for everyone, but given our need for speed, the market is big enough and will only get bigger.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• Infrastructure Q1: IaaS Comes Down to Earth; Big Data Takes Flight
• In Q3, Big Data Meant Big Dollars
• The promise of SDNs in the enterprise

That wasn’t a direct quote, but it’s pretty much the takeaway from Andy Bechtolsheim, the co-founder of Arista Networks (and also of Sun Microsystems).

The combination of low-cost data center infrastructure and rapidly evolving, free or nearly free open-source development tools means that tech startups can get going cheap, and if things don’t work out, move on to other things, Bechtolsheim said today at the GigaOM RoadMap Conference.

When does it make sense for a startup to invest in back-end infrastructure?

Hardly ever, it seems.

A company might build out its own infrastructure only if it’s raised a lot of venture capital, he said. But it needs to be a lot. And even then, maybe AWS is a better way. “Netflix  … uses Amazon for infrastructure. Here’s the leading, largest company in a field deciding it’s cheaper and more efficient to use a competitor for infrastructure rather than building its own.”

Silicon Valley is notorious for its high-cost structure. “It’s hard to justify  the cost of doing anything locally when the cost of power is 30 cents per kWh here vs. 3 cents for Amazon,” he said.

And it’s not just a hardware thing. “The current quality of software tool sets have improved unbelievably,” he said.

“Software programming levels have improved from C to C++ to Java to Ruby to you name it. You can now do more with fewer people and open source deserves all the credit here for creating and maintaining these tools,” he said.

The net effect is that barriers to entry have collapsed for people who know how to use these tools — and who know enough to avoid heavy spending on infrastructure hardware. “That changes the model to allow for experimentation. Unlike ten years ago, when you had to raise tens of millions to get going, now you can do it on your credit card,” Bechtolsheim said.

Bechtolsheim is also on the board of the new Open Compute Foundation, formed by Facebook to propagate specs for standard, energy-efficient data center infrastructure. OCF hopes to bring open-source innovation that so improved software tools into the hardware realm.

For those brave souls wanting to build data centers, the OCF blueprint could help. But, Bechtolsheim said, that’s for truly big companies that need to do huge webscale computing, not for startups.

For nearly every entrepreneur weighing a tech startup, it’s better to rent than to buy or build.

Photo by Pinar Ozger.

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• 12 tech leaders’ resolutions for 2012
• Cloud computing infrastructure: 2012 and beyond
• LTE-Advanced: what it is and isn’t

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• The promise of SDNs in the enterprise
• Cloud computing infrastructure: 2012 and beyond
• The Structure 50: The Top 50 Cloud Innovators

A UBS research note from Wednesday points out that the deal will likely put Dell in closer competition with HP and Cisco. From the note:

Today Dell announced the acquisition of networking company Force10. In the past, Dell resold/OEMed Ethernet Switches from BRCD/JNPR although Dell was not a 10% customer for either. It is likely over time, Dell will prefer Force10 products for Data Center applications. Cisco will also see increased competition from Dell as a result. Lower mgn [margin] hardware companies Dell and HP expanding networking efforts could create greater price competition than in the past.

Dell is moving upstream into networking even as the networking business is facing a shift akin to what happened in servers after the introduction of virtualization. On Tuesday Intel announced the acquisition of a networking chip company that is Intel’s means of capitalizing on that trend. So with Dell moving into this sector with Force10, its relentless focus on the bottom line might make it better suited should switch makers find themselves competing to sell what essentially becomes commoditized networking gear. From the Dell release:

“Today’s datacenter networks are too complex and require too much manual intervention. What worked in the past is no longer viable in the virtual era,” said Brad Anderson, senior vice president, Enterprise Solutions Group, Dell. “Dell’s approach of offering customers open, capable and affordable solutions aligns with Force10’s approach to offering customers new levels of flexibility, performance, scale and automation which is fundamental to changing the economics of datacenter networking.”

It looks like Dell is preparing for the coming era of open networking gear and plans to capitalize on that trend with Force10. For more, check out our virtualized networking panel from Structure 2011 in the video below:

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• Cloud computing infrastructure: 2012 and beyond
• Infrastructure Q1: IaaS Comes Down to Earth; Big Data Takes Flight
• The promise of SDNs in the enterprise

Intel’s decision to buy Fulcrum Microsystems is important and needs to be underscored. Why? Fulcrum makes silicon for Jayshree Ullal’s and Andy Bechtolsheim’s 10 Gigabit, high-performance switch company Arista Networks. The move is a forward-thinking one by the computer chip vendor, as virtualization continues changing the computing landscape. Fulcrum has been around for more than a decade and has raised at least $35 million in venture capital in its 11-year history. From Intel’s release on the acquisition:

“Intel is transforming from a leading server technology company to a comprehensive data center provider that offers computing, storage and networking building blocks,” said Kirk Skaugen, Intel vice president and general manager, Data Center Group. “Fulcrum Microsystems’ switch silicon, already recognized for high performance and low latency, complements Intel’s leading processors and Ethernet controllers, and will deliver our customers new levels of performance and energy efficiency while improving their economics of cloud service delivery.”

Intel’s moves here come amid a shift in the way applications think about architecting their services. Where once a server was an individual computer that ran applications, for cloud computing and large web scale applications such as Facebook, Google or Twitter, they are now components in a much larger system. Google first outlined this shift, but Facebook recently has built upon it with its Open Compute project that basically rethought the way the social network built its hardware and deployed it in a data center and then opened up that design for input and adaptations.

Open Compute left Intel’s (and AMD’s) chip business alone, while putting the margins of systems purveyors such as HP and Dell at risk. However, as other silicon companies such as ARM, Tilera advance, Intel must rethink its value proposition beyond the x86 CPU, much like its customers Dell and HP are now rethinking their value proposition beyond making servers. It’s also looking ahead to the changes that virtualization has wrought in the networking world, where efforts to separate the software that controls where packets go from the switches do the physical routing. This shift is occurring in part because as the data center become a computer, it needs a new way to communicate between nodes (servers). Thus, the rise of fabrics inside next generation servers and inside the data center.

And so back to Intel, which has seen this happening and wants to make sure it has the ability to provide the computing and networking brains inside the new version of the data center. For Intel, the acquisition of Fulcrum enables it to own one more component inside a rapidly commoditizing — but rapidly growing — market. So perhaps we’ll see Intel design a single system on a chip that offers what essentially is a data center on a chip.

Update: Looks like Arista’s Ullal agrees. She emailed GigaOM saying that deal was great for the industry because it validates the concept of merchant silicon as opposed to the specialized chips that Cisco and other networking vendors built. It also moves Intel into the networking sector helping it diversify. Now switch builders can choose from silicon from Intel, Broadcom or Marvell. She added, “This is great for Arista as Fulcrum now has backing of Intel (and truly address the enterprise to cloud networking migration in a mainstream manner). Arista has been one of Fulcrum’s top customers.”

Related research and analysis from GigaOM Pro:
Subscriber content. Sign up for a free trial.

• Cloud computing infrastructure: 2012 and beyond
• Infrastructure Q1: IaaS Comes Down to Earth; Big Data Takes Flight
• In Q3, Big Data Meant Big Dollars