- Stephen Lam/Reuters
And on Wednesday, Facebook upped the bar yet again.
Facebook announced that it was giving away four new designs for brand-new types of computer servers invented at Facebook.
Anyone can take these designs, modify them and use them, with contract manufacturers standing by to build them.
Those contract manufacturers include Chinese companies ike Quanta, as well as the world’s largest maker of computer servers, Hewlett-Packard Enterprise (HPE).
Better and cheaper
While Hewlett-Packard Enterprise has a chip in this Open Compute Project game, OCP’s influence hasn’t necessarily been a boon to the company. That’s because OCP is to data center hardware what Linux is to software: open source. Engineers all collaborate freely on designs, with no intellectual property ownership barriers.
- Thomson Reuters
The idea is simple: to help Facebook and other large-scale Internet cloud companies build faster, better, push-the-envelope hardware that costs less money than buying traditional servers from companies like HPE, Dell or Cisco.
And other big cloud providers, like Google and especially Microsoft, have been involved in OCP, using designs themselves and sharing some of their own tech innovations.
Apple, too, is involved, as is AT&T, Verizon, Goldman Sachs, Bank of America, and many others. Meanwhile, Amazon and LinkedIn are also building their own infrastructure from scratch, if not from OCP designs, then from their own.
As businesses choose to use cloud-computing services instead of buying their own servers, and the internet companies build their own faster, cheaper servers, the server industry has begun to hurt. In 2016, revenue for the worldwide server market declined about 4% to just under $53 billion, according to market researcher IDC, in large part because of cloud computing and “declining high-end server sales,” said IDC research director Kuba Stolarski.
This includes double-digit declines in revenue in the last half of 2016 for HPE.
Raising the bar
In many ways, Facebook’s designs have been putting the professional hardware industry to shame, doing things with hardware that’s never been done before.
- LinkedIn/Arlene Gabriana Murillo
That’s because people watch 100 million hours of video a day onFacebook; they post over 95 million photos and videos to Facebook and Instagram daily, and a good 400 million people now use voice and video chat every month on Messenger, Facebook tells us.
That kinds of usage puts incredible demands on the computer infrastructure. Facebook has been inventing new tech in response, and then giving away what it creates to OCP.
To that end, Facebook’s technical program manager Arlene Gabriana Murillo announced these four new servers on Wednesday, which adds to and updates a handful of other servers it has already contributed to OCP. Take a look:
“Bryce Canyon” is a storage server for things like photos and videos that can hold more (20% higher hard disk drive density) and run faster (4x increase in compute capability) over the predecessor Facebook invented (known as “Honey Badger”).
“Yosemite v2” is a server that keeps chugging away even when someone yanks out bits and pieces of its insides to replace or repair them.
“Tioga Pass” is a computer server that processes big chunks of data in memory, making it faster than conventional servers as well as its predecessor “Leopard.” (It uses dual-socket motherboards and more i/o bandwidth for things like memory and graphics processing).
“Big Basin” may be the most interesting one of all. It is a server used to train neural networks, the “machine learning” technology. Artificially intelligent computers must consume large numbers of examples of something in order to recognize it. For instance, they must see gobs of photos of dogs in order to learn that a German Shepherd and a Chihuahua are both labeled “dog.” Big Basin can consume 30% larger data sets than the previous training server (named Big Sur), making AI systems smarter and more reliable.