Facebook leads open computing and data center giants row

Open computing is a new concept proposed by http://www.aliyun.com/zixun/aggregation/1560.html ">facebook in 2011, so far the open computing project is becoming a force to be reckoned with. The giants of several data center industries are actively committed to supporting and developing new products in accordance with the personalized specifications Facebook requires.

In the fourth session of the Open Computing Summit in Santa Clara City, California, this week, Intel and AMD are publishing new products based on open computing Projects (OCP). OCP is a design specification for Facebook's server and data center, which was open 18 months ago.

Under the supervision of financial services companies Fidelity and Goldman Sachs, competitors AMD and Intel have developed a motherboard model for OCP open rack architectures and standard rack environments.

The idea of the motherboard model was first presented at the first Open-ended Computing summit in October 2011. At that time, AMD and Intel engineers met with financial services executives, who described their strong interest in open computing hardware, but had to combine them with their existing data centers. "They tell us that they are currently under the constraints of the existing architecture, so they want to be able to see open computing hardware that fits their existing rack architecture, because they are not able to splurge on their data centers," recalls Bob Ogrey, a AMD colleague and cloud computing technology dedicated to the server platform architecture.

The resulting project, AMD's Roadrunner motherboard series and Intel's decathlete motherboard series, are suitable for early adopters and will be ready for production this quarter.

Further, depending on the server, network, and management options, AMD's motherboards can be configured for cloud computing, high-performance computing, or storage servers. In addition to the existing grid computing clusters, it is important for financial services organizations that attempt to reduce computational costs to build large internal private clouds. "They want to be able to build cloud computing and grid computing on a single platform," Ogrey said.

Open Computing in large environment

Indeed, given the summit, open computing design is no longer limited to Facebook's own data centers. When cloud computing provider Rackspace claims to personalize the OCP design in the current environment, the game provider Riot Games says it will buy Hyve servers produced by the original design manufacturer Soltions OCP.

Rackspace, for example, borrows from the open computing open rack specifications of Facebook's original design, and is transformed according to the realities of its own data center. The resulting design features include higher peak power supplies, abandoning DC power, supporting traditional switches and adding a cable management bracket.

Working with Quanta,delta and Wiwynn, the OCP design for the production of modified products has been able to achieve the goal of reducing costs and accelerating time-to-market, Rackspace chief Operating Officer (COO), Mark Roenigk said at a meeting. Compared with existing designs, capital expenditure can save about 40%, and operating costs are expected to save 50% to 52%. These savings, he added, are quite significant, "but it is equally important to quickly improve the speed with which we deploy our systems." ”

An ecosystem derived from open computing

The scope of the OCP is also extended. To date, this program has been found to be applied to servers, rack design (open racks), storage (open storage), power and hardware management. Recently, the group has set up a new compliance and interoperability team to test and certify the OCP program.

At the summit, Intel also pledged to design for the upcoming silicon Photon technology. This is a network technology that can support 100Gbps switching in a rack.

But OCP ultimately, the Facebook leadership is still responsible. New contributions include: A new, full Flash database server ("Dragonstone") and the latest version of the "Winterfell" Web server in a cold storage environment for open racks and open storage. Facebook has also contributed to their "Group Hug" motherboards, a motherboard that provides common slot architecture specifications, which can hold up to 10 different vendors of system-level chips (systems-on-a-chip), or SoCs.

The problem of missing a common slot has plagued hardware designers for a long time, said Frank Frankovsky, head of Facebook's hardware design and supply chain. "We can't design a common socket that often makes me suffer." ”

(editor: Heritage)