Open Compute Gains Switch OS, Storage, Rack Contributions

The Open Compute Project took several steps toward its goal of creating a set of building blocks for the modern data center with advances announced at its 2016 US Summit.

Charles Babcock, Editor at Large, Cloud

March 11, 2016

7 Min Read
<p align="left">(Image: Open Compute Project)</p>

10 Startups Aiming To Change The Way You Work

10 Startups Aiming To Change The Way You Work


10 Startups Aiming To Change The Way You Work (Click image for larger view and slideshow.)

The Open Compute Project, previously limited to a few, innovative motherboards in the public domain, is taking on more depth and variety. OCP now encompasses solid-state disks, an operating system for generic switches, and equipment racks of different designs -- including one backed up by a novel 48-volt power supply.

The latter was donated by OCP's newest member, Google, which designed its own racks based on its experience building its search data centers. Most battery backup power supplies are based on 12-volt designs. Facebook, which founded the Open Compute Project (OCP) in 2011, uses battery packs delivering 12 volts to the rack.

By keeping the voltage at 48, Google loses less power as it's delivered to each shelf in the rack, where it is then stepped down to 12 volts and converted to the alternating current that the devices can use. The higher the voltage, the less power lost in transmission and conversion to alternating current.

Google waits until the last step, feeding 48 volts up to the motherboard's entry point, where the conversion is made, instead of doing it at the top or bottom of the rack. Delivering 48 volts to the device level "reduces energy losses by 30%," noted Urs Holzle, Google's noted chief infrastructure architect, during an appearance at the opening general session of the OCP US Summit in San Jose, Calif., on March 9.

The batteries serve as the emergency backup power that keeps everything running for 12 to 15 minutes in the event of a power outage on the grid. That is presumed to be more than enough time to get emergency generators onsite running and cut in to supply replacement power.

Holzle said Google has also come up with a new rack design, one that is shallower than the standard racks used in most data centers. That's because it has already packed its data centers full of equipment of the standard depth, and "our rows are too shallow" to fit any more in. A new rack design, however, would regain space for additional equipment.

Asked whether Facebook will also move to a 48-volt power supply, a spokesman said such a migration would require the redesign of its existing racks, and for the time being Facebook is satisfied with its own power-saving measures.

Learn to integrate the cloud into legacy systems and new initiatives. Attend the Cloud Connect Track at Interop Las Vegas, May 2-6. Register now!

If anyone doubts that the OCP represents the future of data center design, Holzle's appearance should diminish those doubts. Google has long been regarded as a well-heeled innovator in data center design. It is believed to be the first Web company to design its own servers instead of buying from existing models. It came up with a modular server with duplicate fans, power supplies, and other redundant parts stripped off its chassis. The Google server was a partially enclosed motherboard that was also easy to install and uninstall in a rack.

Holzle's appearance at the OCP Summit means that Google, Facebook, Intel, HP, Cisco, Microsoft, and a broad set of financial services and telecommunications companies are behind OCP, along with many smaller equipment manufacturers. With many of the most-advanced data center builders contributing to its "open source" equipment set, it's likely to draw increased scrutiny from former skeptics.

Equinix, a major data center operator with compute/communication facilities in many metropolitan markets, is another new member. It joined OCP to cooperate with Facebook on a joint design for the Wedge 100 switch, a unit advanced enough for Equinix to plan to install it in many of its 145 data centers, CTO Ihab Tarazi told InformationWeek in an interview.

In addition, Microsoft's Azure Cloud CTO Mark Russinovich presented OCP with the SONiC switch operating system, which can be used with the generic switch designs coming out of the project. Russinovich also spoke at the Summit's opening. He confirmed that SONiC runs on top of Debian Linux and that it will serve as the device operating system that accepts instructions from a controller. (This is most likely the first Microsoft device operating system based on Linux.)

[Want to learn more about Microsoft's investment in OCP switch software? Read Microsoft Leads Open Switch Software Effort.]

OCP is coming up with hardware and software with which to build software-defined networks that can have one set of characteristics at the start of the day, and then revise them as traffic or business needs change. Changing traffic patterns could trigger policies governing the network that tell the controller to issue new instructions to the switches and routers to shift how they're handling the traffic.

Russinovich said SONiC is likely to be the operating system on OCP-based switches, because it's the only one that's been submitted so far. "Most of the other elements were in place [to build an OCP data center network]. This was the last missing piece," he said in an interview. So far, Microsoft has submitted the OS, and OCP's Network Project working group will need to review and accept it.

Intel's Jason Waxman, a member of the Open Compute Project Foundation board and corporate VP of the Data Center Group, told the opening session that Intel has donated the design for a solid-state storage system, dubbed Lightning. Lightning was codesigned with Facebook and is based on non-volatile memory chips. It comes in the form of a 2u sled loaded with solid state disks for an OCP server rack.

Waxman said Lightning is "a next-generation data center storage architecture" that can be used as a large storage pool to accelerate the performance of storage-oriented applications. It can be used with Internet of Things applications, big data analysis systems, machine learning, and other applications that require large amounts of data and high I/O throughput.

A sled of Lightning taking up 2u in an OCP rack represents 120 TBs of storage, said Intel's Raejeanne Skillern, VP of the Data Center Group, in an interview. Thirty solid state disks, each with 4 TBs of memory, are packed into the sled.

Skillern said non-volatile memory, including that produced by Intel, is useful in obtaining quick reads and writes for modern applications. Data travels over a high speed PCI Express data plane within the device. "We're bringing more storage into the rack instead of providing it as a separate storage system," such as a storage area network or a network attached storage device. Lightning will provide six times the I/O operations performance and half the retrieval latency of existing SATA SSDs, said Skillern.

These contributions represent steps toward achieving the basic goals of OCP, said Jason Taylor, chairman of the OCP Foundation and VP of infrastructure at Facebook. It set out to create "a small number of powerful building blocks" on which a modern data center could be built. The building blocks would be standardized, and they could be produced by multiple manufacturers from a common set of designs.

Both the open design and multiple sourcing would "avoid vendor lock-in" and allow the next generation of cloud suppliers and participants in the digital economy to avoid investing in "pointless differentiation," he told attendees at the March 9 general session.

The two also represent a spreading out beyond Facebook's own data center innovations to those of other companies, and a move beyond core motherboard servers into much of the supporting infrastructure, such as virtualized networking and storage.

About the Author(s)

Charles Babcock

Editor at Large, Cloud

Charles Babcock is an editor-at-large for InformationWeek and author of Management Strategies for the Cloud Revolution, a McGraw-Hill book. He is the former editor-in-chief of Digital News, former software editor of Computerworld and former technology editor of Interactive Week. He is a graduate of Syracuse University where he obtained a bachelor's degree in journalism. He joined the publication in 2003.

Never Miss a Beat: Get a snapshot of the issues affecting the IT industry straight to your inbox.

You May Also Like


More Insights