The Open Compute Project (OCP) started off as a small team taking on a massive project that answered the question: how do we scale the computing infrastructure at Facebook in an efficient and economical manner? This three-person team was attempting to overhaul data center design. They designed their own servers, power supplies, server racks, and battery backups. Decades-old designs and standards were tossed to the curb in search of solutions to power the backbone to one of the world’s largest data gathering companies.
For reference, Zephoria reports that 1.35 billion Facebook users upload over 300 million photos each day. That’s up from 200 million just two years earlier. The infrastructure to handle this amount of data and subsequent growth (5 new users sign up each second) is what lead to the creation of the OCP team in search of new data center designs.
The team’s work culminated with a completed data center located in Prineville, OR in 2011, which employs 126 people today and uses 38% less energy to do the same work as existing facilities while costing 24% less to operate. Much of that cost and energy savings comes from a design that gets rid of air-conditioning system that are used to cool most traditional data centers.
The team then did something you don’t often see if the highly competitive data center market: they shared their designs and best practices with anyone who was interested. In short, they open sourced the data center.
“We are not the only ones who need the kind of hardware that we are building out,” said Mark Zuckerberg, the co-founder and CEO of Facebook, in a piece by the NY Tmies. Facebook’s goal with OCP is to encourage discussion and collaboration for future data center designs and improve upon the work they accomplished at Prineville. Today the project is open to others with similar goals.
While still relatively new, OCP is already strutting its influence on both hardware and software design. Let’s take a look at a few area in which OCP might flex its muscle in terms of influence.
The OCP is setup as a non-profit in the state of Delaware and has seven individuals serving on the board of directors. These leaders come from companies such as Intel, Microsoft, Rackspace, and Goldman Sachs. Bringing this level of experience and resources to a project allows companies to focus on their area of expertise while reaping the benefits of the group’s collective innovations. Microsoft can focus on creating software services and tools while Rackspace shares what they’ve learned from building out large data farms. This doesn’t mean there won’t be competition among the group at times. For example, AMD and Intel will continue to fight to entice customers to its version of the server compute node that’s one of several OCP projects in the works. But companies will be able to solve problems that have already been solved through collective knowledge sharing, which frees up resources to focus on their specialty.
Customer Driven Design
If you think about it, the small group at Facebook went off searching for new solutions because they were not satisfied with the current models. Those models were often locked-in designs controlled by the hardware makers rather than the people who design and manage the centers.
With OCP designs, the process of creating new products isn’t locked to any one or two companies. At Puget Systems, we’ve recently experienced what happens when motherboard designs are controlled by a handful of companies, and it’s not pretty. Choice goes in the tank, costs go up and stability goes down. If you’re the only game in town, you aren’t all that incentivized to listen to your customers.
OCP promises that innovative ideas will be brought to market faster, and quickly be tested under some of the most demanding situations. Those that pass the test could be well on their way to more mainstream success. OCP hopes that providing a level playing field that more companies will want to contribute and innovate.
Beyond the PUE
Efficiency is measured a number a ways, but in the data center world, it’s often the PUE (Power Usage Effectiveness) that garners the lion’s share of attention. But efficiency also includes looking at how effectively a product is designed, what materials it uses, and how much it weighs. It’s the difference between focusing on a handful of metrics and taking a holistic approach to creating functional designs that can be replicated using materials and resources in the most effective fashion.
Joel Wineland refers to this type of design as “vanity free” with a focus on providing value over performance at any cost. Data center build-outs tend to focus on the technology that resides in these vast spaces. A holistic and inclusive approach to design can find efficiencies and savings in areas where little attention has been given, even down to how servers are racked and powered.
OCP hasn’t been without its detractors. When Facebook announced the project Eric Knorr from InfoWorld wrote that OCP was an example of Facebook exercising its sizeable leverage over OEMs to drive lower prices and encourage design in those areas that benefit Facebook the most such as servers and storage. There’s no doubt that Facebook wanted its hardware partners to strip away the unnecessary features thus reducing costs, power consumption and footprint.
But there’s little doubt today that large data centers do have specific needs, and that the tiny details add up when building out a date center on such a large scale. Facebook has massive computing needs and is in a position to get what it needs, even if that means sharing its ideas with competitors.
Do you believe that Facebook was wise to create the OCP and will benefit from the collective knowledge of the group and competing OEMs?