Intelligent Infrastructure

Mass data and next-gen workloads Intelligent Infrastructure

Facebook optimizes cold storage for better performance

Facebook's focus on data center efficiency could provide valuable lessons for cloud vendors as the need for optimization grows.

While Facebook is a consumer-driven business, it has offered several lessons in data center design through the Open Compute Project that can service as a model for many companies. The social network also makes for an interesting case study because of the sheer volume of information that its cloud hardware must accommodate. Data Center Knowledge Editor-in-Chief Rich Miller went through an extensive investigation of Facebook’s data ecosystem, revealing some of the factors that have contributed to its success in creating efficient infrastructure.

In addressing the issue of scope, Miller noted that Facebook currently stores more than 240 billion photos, with 350 million new images uploaded on a daily basis. This translates to close to seven petabytes of new storage capacity each month. In addition, the content viewing habits of is users have created challenges that extend beyond the issue of sheer volume. An estimated 82 percent of the site’s traffic is concentrated on 8 percent of all hosted images. One of the common solutions to this type of problem is tiered storage, in which information is categorized and moved to different cloud storage tiers depending on how much performance is required.

Facebook’s cold storage solution
The traditional option of static hot and cold storage would not work for Facebook. The company faces a unique challenge in that it needed the ability to deliver rarely accessed data quickly. It would be easy to imagine the end-user frustrations that would arise if a consumer couldn’t easily access their childhood pictures at a family reunion, for instance. Miller reported that Facebook developed software to solve this problem. The program can categorize and dynamically move data across three storage tiers in response to demand.

Adding to the efficiency of the solution, Facebook housed its cold storage at its 62,000 square foot data center in Prineville, Oregon. Miller noted that the facility can house 500 racks that add to the total of one exabyte of cold storage capacity. The company plans to build similar facilities in North Carolina and Sweden to expand on those efficiency gains.

“The cold storage data center has no generators or uninterruptible power supply (UPS), with all redundancy handled at the software level,” Miller wrote. ” It also uses computer room air conditioners (CRACs) instead of the penthouse-style free cooling system employed in the adjacent production data centers in Prineville.”

Bigger data needs
Most businesses know that they must accommodate growing quantities of information, but the era of much larger data may arrive sooner than many think. Analysts have predicted continued momentum for the cloud market, and there is already a significant amount of information in the enterprise cloud storage space. Symantec’s 2012 State of Information Report: Digital Information Index revealed that 46 percent of medium and large business information worldwide is stored outside the corporate firewall, with cloud storage holding 23 percent of all corporate data.

Cloud adoption is well-positioned to accelerate over the next 12 months, according to research firm IDC. In late 2012, the company predicted that 2013 will be largely dominated by advances in mobile and cloud technology. Worldwide IT spending is expected to increase 5.7 percent from 2012 to the end of this year with mobile leading at 20 percent growth. The cloud is also expected to play a significant role in driving spending as more vendors will make the technology a central part of their service portfolios.

As more businesses incorporate cloud technology into their IT optimization strategies, cloud storage companies will also need to place a premium on making the most of their resources. This will likely necessitate a mixture of paradigm shifts, such as Facebook’s new data management strategy and the implementation of highly efficient hardware so cost savings can be achieved without negatively impacting performance.