Why you can't ignore the Open Compute Project any longer

If you’re in the data-center business, you’ve probably been following the Open Compute Project (OCP). And if you haven’t, you should. OCP is Facebook’s response to dealing with a server count that likely exceeds a million. Their innovations around scale and serviceability are simply too compelling to ignore.

The industry can’t ignore the designs rising out of OCP. It’s a very disruptive approach. I’ve been working in data centers since a 300 MB disk was the size of a washing machine, and I’ve watched what happens when companies are stubborn to adapt. You’ve got two choices when disruption comes to the marketplace: embrace it or die a slow, agonizing death.

The OCP naysayers want to dismiss these new designs. I hear people say that you have to be the size of Facebook for it to make a difference, or that most companies can’t afford to adopt OCP. I don’t believe that’s true. Companies do not have to be at Facebook scale to see the benefits of Open Compute designs. It’s not just about rack costs, it’s also about manageability and total cost of ownership.

Cost vs. power savings

OCP racks might appear to cost more. But when you factor in the savings in power along with reduced server spend on power supplies (which are external to the servers in a shared configuration per rack), along with the opportunity to move the UPS to the rack (lower blast radius for failures and opportunity to virtualize power), the true cost of ownership for OCP is less than that of traditional server configurations.

What does OCP mean for data centers? It’s disruptive. That’s why people are a little afraid of it. If you’ve bought your servers from traditional legacy vendors, you get stuck in a walled garden. You are directed to purchase hardware with features you don’t likely need and expensive service plans. Along comes OCP, giving you more self-service options, denser configurations, and stripped-down motherboards. Your IT loads are still critical. But smarter software is what will drive your resiliency — not two power supplies per server, an over-cooled data center, or pricey support from an original-equipment manufacturer’s field-service team.

Enterprise IT departments and other providers need to start treating the hardware like the commodity it’s become. You no longer need to baby the machines and pay for extensive field service plans. Disks, dual in-line memory modules (DIMMs), power supplies and even motherboards are all replaceable by your own staff on most systems. OCP makes that even easier by designing all field replaceable units to be snapped in and out with just your hands. Focus your energy on service that offers parts only, at market rates.

Rethinking server designs: ugly is the new pretty

Traditional server designs are these monolithic machines, emblazoned with big logos, fancy facades — they’re pretty as hell. OCP design takes away the bezels, takes away those fancy facades. Why? They block airflow. The minimalist design keeps everything cooler, using less energy per work unit, pushing toward ultra-low power usage effectiveness (PUE). Stripping out the excess skinnies down the server size. Normally, today you would have a rack full of servers, up to 40 one-rack-unit pizza-box-style servers right on top of each other. OCP racks can fit three servers side by side in a two-rack-unit height, so you can fit 60 in the same height. That’s incredibly dense, and for a nerd, it’s sexy as hell — bare bones, running hot and noisy. At a glance, you might think, “Man, those servers are scary ugly!” But when you think about it, they’re beautiful. If efficiency was the bass, they’re all about the bass and no treble. While I know it’s more than OCP that drives Facebook’s low PUE, when I hear about PUE under 1.1, I’m just jealous. I know OCP designs are a big piece of that savings.

Ditch the downtime

But you have to think about power supply. Downtime is the fear of the data center. Traditionally, you run two power supplies per server. If one fails, the backup is always there. It’s extremely wasteful, though. Imagine driving around with two engines in your car on the off-chance one choked up. You’re just burning fuel. OCP designs share power supply for the entire rack. Instead of 80 supplies for 40 servers, you have six or so larger supplies. They run more efficiently, and maybe you need a seventh or eighth for an N+1 or N+2 configuration. OCP design puts batteries right in the rack, eliminating the need for data center sized uninterruptible power supplies. This also means you can control the power supplies with software and turn off those supplies that aren’t in use. If a supply fails, the local batteries cover the power, while a new supply comes online in your N+1 configuration and a call goes out to your techs, telling them which power supply to replace. Less components (power supplies) and lower power consumption with the same availability allows you to run more servers for the same watts and not sacrifice availability. What’s not to like?

Acknowledge tradition, but embrace the future

To run a modern data center, understanding and embracing OCP is a must. Does that mean we’re telling our customers to clear out their traditional racks? No, we still have an investment in traditional technologies. It’s really about flexibility. It means we have an expanded space for OCP designs. That’s both literal and a mindset. We’re not interested in saying, “My daddy built his data center this way, and his daddy’s daddy before him.” If customers come to us and say, “I’d like to go OCP,” we’re excited to make that happen.

All disruption comes with obligatory fear: What’s this going to do? What’s the impact? Eventually, the cost difference will be so attractive that people will get over the fear and adopt. OCP will continue to gain legs. Even if it seems like a slow start, it’s not that slow. We can’t ignore it. I’ll definitely be paying attention.

Chris Yetman is senior vice president of operations at Vantage Data Centers, where he is responsible for leading operations, security, network and IT. He has more than 25 years of operations, engineering and IT experience in the Internet infrastructure industry. He previously served as vice president of infrastructure operations at Amazon Web Services, where he had worldwide responsibility for operations and network for Amazon’s data centers.