Sometimes technology stagnates and markets don’t give us what we want. What can we do when this happens? Demand disruption! As a buyer we can insist that our suppliers form a collaborative, idea-sharing business ecosystem—with their direct rivals. We can insist that those who are central to the current industry do what they can to lower barriers to entry for newcomers. We can ask unthinkable things of our suppliers. We can demand better.
Facebook, Applied Micro, Calxeda and Red Hat
Frank Frankovsky is head of hardware at Facebook. He is a big, bearded man who lives in Austin, Texas. He spends a good deal of time at Facebook headquarters in Menlo Park, California. He has problems with his server farms — well, to start with he’d rather they not be either “farms” or contain conventional “servers.” He wants to “break up the monolith” of server architecture — so that fast-evolving parts such as microprocessors are interchangeable, while slower-moving parts such as communications channels are only changed as needed.
Frankovsky is initiating a business ecosystem called the Open Compute Project.[i] He has a vision of where he wants to go and how he wants to measure success. His current thinking is summed up in “Group Hug” — which includes taking what is today on a motherboard — which evolves slowly — and standardizing its functions and its interfaces. This will in turn enable microprocessors to be switched out when they become obsolete — which is quickly — without replacing other perfectly adequate parts of the system.
Frank has invited a number of companies to the table, and many others have shown up — at least 50 companies smell a rich opportunity. Frank wants it this way, he wants to encourage investment by coming as close as possible to assuring that if they build it he will buy. He ideally wants more than just better products. He would like the companies he is involved with to collaborate and move toward open ecosystem models.
Frank is exploring how far he can go toward an open source collaboration model for hardware companies. He points out that the current model for open source communities is suited for software, because individuals can make substantial contributions as short-term volunteers.
“In an open software ecosystem an individual’s passion and an individual’s skill and talent is almost all you need. A software engineer can change the world with a few nights of coding. Their contribution is their intellectual property and their time.
“Whereas in the hardware ecosystem we have tangible infrastructure that needs to be built and that requires tangible dollars.”
He and I share an interest in hybrid ecosystems that combine open and commercial models, such as the ARM ecosystem.
“We are still exploring what open source looks like in the hardware ecosystem. How open can it be, how collaborative can it be and where does there need to be more traditional property rights so that people who engage in the ecosystem can build a reasonable commercial opportunity out of this?
“This is an area that fascinates me. We hope to break down some barriers with Open Compute.
“When we established the IP policies we leveraged the Open Web Foundation agreements—clearly oriented towards software.
“But then one of the founding members of Open Compute is Intel.”
Frank went on to say how Open Compute and Intel had worked together on a model copyright agreement so that Intel — and others — could contribute specific intellectual property while drawing a boundary around it. Then the agreement was shared with other members, and because Intel’s legal department had agreed to it, other companies were inclined to believe it would protect their interests.
“It’s still far from frictionless but we are seeing some amount of change. I will give you a great example. Intel has invested for more than a decade in the silicone photonics technology.
“Thinking and talking about open sourcing that technology was a very interesting discussion. I don’t want to speak for them but from my perspective the reason that they came forward what because they saw the benefit of how quickly Open Compute could make that technology pervasive.”
I noted that, counter-intuitively, Frank is not asking his suppliers to lower their prices or to make better products, he is asking them to open up to a new model of sharing. And he is trying to convince them that this is in their best interests — for example, to open up in order to have their technology proliferate.
“They are not open sourcing the process technology that allows them to provide photonics and silicone, which is their core intellectual property.
“But they open sourced the connector so that anybody in the ecosystem can download the specifications and develop a product around the photonics.
“So that’s a sign of things to come. Intel obviously has a lot to protect. And even they are thinking differently about this ecosystem. I think they see that the world is changing and business models are changing.”[ii]
By early 2013 four companies had built and demonstrated hardware that allowed processors to be supported interchangeably — Intel, AMD — both X86 architecture — and AppliedMicro and Calxeda, representing ARM architecture. Intel engineers have come up with a socket design that is processor agnostic, and all four are using it. What this means is that as Facebook continues to advance toward implementing its new server architecture it will start to put ARM vendors on a precisely equivalent hardware footing with Intel and AMD.
Why would Facebook do this? While continuing to work with Intel and AMD, Facebook wants access to the diversity of the full connected community and its ecosystems. Today Facebook has two new suppliers, AppliedMicro and Calxeda, tomorrow it will have more, the next day even more. Frank is holding open the door to invite new ideas and talent into his data centers.
The ever-differentiated ARM ecosystem has in Applied Micro and Calxeda two very different approaches to a server processor — two different bets. Both will be made available to Facebook. Applied Micro has an ARM architecture license and is making an ultra-high-performance processor. In raw performance it is designed to roughly equal a top Intel processor, but be more power-efficient.
Applied Micro is headquartered in Sunnyvale the Silicon Valley, a dozen miles south of Facebook’s headquarters in Menlo Park. Paramesh Gopi is the colorful CEO of Applied Micro. He equated his goal to building the BMW of ARM chips,
“What we’re driving toward is optimal performance—the ultimate driving machine.”
Picking up the bait, I said,
“So you’re a BMW, not a Prius.”
“Exactly, I think we should talk about us more as a Lexus 450 hybrid —or a Lexus GS hybrid.”[iii]
That is, a luxury car with plenty of power and interior passenger space, and with the ability to run on electric power when appropriate. Later he spends several minutes explaining that in order to succeed in the large-scale server and data center market, ARM partners will have to match the cadence of the existing cloud data center ecosystem. Paramesh notes that his offering is intended to do just that, to be an easy upgrade, and once having entered the cloud world Applied Micro will add further differentiation away from Intel and AMD.
Meanwhile from their headquarters in silvery cylinder on the west side of Austin, Texas, Calxeda’s Karl Freund explained their approach.
“We don’t want to be almost as good as an Intel Xeon, we want to be different.”
Different in the Calxeda case is putting together many ARM processors tied together by a special energy-efficient connecting fabric. They see themselves at the beginning of their evolution; right now Calxeda chips can handle light jobs such as file serving, and do it power-efficiently. The fabric and the processors come as a unit that can be made interchangeable with the most common Intel chips used in data centers like Facebook.
“Customers like us because of our open business model and ARM’s open model.”
At the highest level, Calxeda’s promise is that they will innovate for and with the customer and that they have access to solution elements that are potentially powerful. Karl goes as far as to tell customers,
“We are probably not the only choice for you. You may decide to buy from us for certain applications and buy from Applied Micro for others. What we can assure you is that in either case the software will be compatible. Once you set yourself up to use any ARM systems, the diversified ARM ecosystem will continue to expand the options available to you.”[iv]
A couple weeks later I stepped into a funky Cambridge, Massachusetts coffee house a few blocks from MIT. Jon Masters of Red Hat was already there. He’s a regular, and he was chatting up the person behind the counter as she made his coffee. We tucked ourselves into the back corner booth. Jon is an ultra-high energy computer scientist who is the chief ARM architect at Red Hat, the open source Linux company. We talk about Facebook, Open Compute and the two ARM companies.
“Both companies are taking different approaches, [Applied Micro] are doing a custom design — very fast, very high end 64‑bit Encore in the next technology generation. Calxeda are going to also have a 64‑bit design but the difference is Calxeda are not doing a custom core design. Their value is about integrating fabric across the chips.”
Red Hat will support both, though it currently only supports ARM through its experimental Fedora version.
For Red Hat, an open ecosystem is a mixed blessing. It currently can focus on a single architecture. This simplifies business. On the other hand, he sees how this simplification has held back progress.
“We’ve [the IT space] had a convenience in the last decade or so where we’ve had certain large players that have had a strategic stronghold in the market. And good for them, they’ve done great. But lack of competition is why we’re looking at things like hyper scale computing now and not five years ago. Fast-forward five or ten years from now you’re going to see a much more dynamic marketplace.”[v]
Jon is enthusiastic about Frank’s initiative because it is accelerating the opening.
Stepping back from AppliedMicro and Calxeda, one can see how effective Frankovsky’s ecosystem initiative may be. It already has managed to press Intel and the three other firms to agree to make their processors interchangeable. It has added two creative and very different voices to the table. It is establishing communication among companies that would not normally talk, and it is forcing each one to pay attention to the others.
Facebook is an example of a powerful customer insisting that its vendors form an ecosystem around its priorities. It is not alone. Companies like Fidelity Investments are also looking for custom data center designs. A highly differentiated ecosystem with lots of companies will find a way to serve them. And on the telecommunications side, large carriers like AT&T and BT are joining a Network Functions Virtualization[vi] project to encourage ecosystems for software defined networking, so that infrastructure is open and interchangeable.
The promotion of business ecosystems by large, powerful buyers almost guarantees disruptive results if the talent can be held together, because it encourages new people with new ideas, new money, new tools and new technologies. Customers give the newcomers precious information about their problems and they promise a market. I look forward to staying tuned to these stories.
More generally there is a theme here that I’d like to apply to my own professional and personal life. I can ask, “How can I work with those who supply me to encourage them to become open to new ideas and new contributors, so that in the long run I’m served by a more vibrant, more diverse and multiparty ecosystem?”
We all put up with things that make no sense, in both our professional and personal lives. Rather than putting up with it, can we use our buying power to encourage open ecological change? If our own buying power and power of persuasion is not enough, can we band together with still others to increase our collective clout? Can we form customer-led, change-making ecosystems? Can we establish ecosystems to promote open ecosystems? I’m inspired that Frank Frankovsky at Facebook is doing this and having some effect and learning a great deal.
[i] Open Compute Project, http://www.opencompute.org
[ii] Frank Frankovsky, Facebook, personal communication, February 2013
[iii] Paramesh Gopi, Applied Micro, personal communication, February 2013
[iv] Karl Freund, Calxeda, personal communication, February 2013
[v] Jon Masters, Red Hat, personal communication, March 2013
[vi] “Leading operators create ETSI standards group for network functions virtualization,” ETSI, January 22, 2013 http://www.etsi.org/news-events/news/644-2013-01-isg-nfv-created