Facebook, one of the driving forces behind open computing, this week updated its designs for an open storage chassis, an open server, and several other open compute platforms.
They are freely available to anyone, through the Open Compute Project (OCP). The announcement was one of the caps to the annual OCP conference in Santa Clara this week.
Demands on data centers are forever increasing. In Facebook’s case, the company is seeing an inexorable, rapid increase in the use of photos and videos that it has to keep up with. So, although Facebook wants open hardware, that hardware is getting progressively more complex. The upgrades announced this week are all either denser, faster, more powerful, more configurable, or some combination thereof.
Facebook said it plans to refresh its entire server hardware fleet with equipment based on the updated designs.
“As photos and videos become even more central to the way people connect and share with each other, the efficiencies we gain help us scale and improve the speed of our infrastructure,” a Facebook spokesman said in an email exchange.
Asked if Facebook is intimating a new wave of spending to get the new hardware installed on some accelerated schedule, a Facebook spokesman explained to SDxCentral that the company started work on the refresh more than a year ago. The company continuously updates its hardware and the update schedule hasn’t changed.
Facebook’s new Bryce Canyon storage chassis will improve upon its predecessor, Open Vault, in several ways. Among them, it will support 20 percent more HDD (high density drive) capacity, four times the memory footprint, and have better thermal performance – a critical consideration.
Another update is Big Basin, the successor to the company’s Big Sur GPU server, a specialized bit of equipment that lacks general compute and networking capabilities and so must work in tandem with another server. Facebook uses this class of GPU servers to train neural networks. The company said Big Basin can train models that are 30 percent larger than Big Sur. It can also render impressive throughput increases. The practical result is the ability to tackle more complex problems and get results faster.
Tioga Pass is the successor to the company’s Leopard, a server Facebook said it uses for several tasks, including working with the Big Basin GPU server. The new spec allows for more options for setting up memory and compute resources, so that it can be configured for different applications.
Yosemite v2 is a refresh of Yosemite, the company’s first-generation multi-node compute platform. The design still holds four single socket (1S) server cards, but previously, if one card needed to be swapped out, all four had to be turned off too. One of the key advantages of the new design is that the other three cards can remain operating while the fourth is swapped out.