LinkedIn Connects With a New Data CenterBy Mike Vizard | Posted 2016-04-26 Email Print
LinkedIn must develop new applications for its members that will drive higher margin revenues. To achieve that goal, the firm is adding a megascale data center.
At the core of that architecture is an existing LinkedIn pod configuration that is employed across all four of the company’s existing data centers, which are located in Texas, California, Virginia and Singapore. Each of those pods can be configured with thousands of servers per pod, and a total of 64 pods can be connected across a flat network fabric.
All the switches used on those networks are identical and fit into a standard 1RU rack. In time, the network fabric and server architecture LinkedIn created for use in the new Portland facility will also be applied to those four existing data centers.
LinkedIn considered all of its IT infrastructure options before deciding on the Portland facility, including using public cloud services. But the cost of using those services at the level of scale required by LinkedIn would not have been financially viable, according to Bachar. The model chosen allows LinkedIn to defer construction costs by leasing the buildings, while investing in an innovative network fabric that takes advantage of the economics of white-box servers to reduce capital expenses.
Colm Keegan, a senior industry analyst with Enterprise Strategy Group, says that despite all the hype, there is no "perfect size" IT model that fits every organization. In the case of LinkedIn, the size of the IT investment makes it less expensive to customize the IT infrastructure in a leased facility. Other companies are relying on a mix of public and cloud services, while others still prefer to focus only on their core business.
“These days, everything is on the table,” Keegan says. “It’s really about how core IT is to your business model.”
Containing Operational Costs
In the meantime, the biggest challenge that remains is containing the IT operational costs that still consume the vast majority of the typical IT budget.
Bachar notes that the pods LinkedIn created make it possible for IT infrastructure elements to fail in a way that doesn’t impact the overall IT environment. Each failed component is then replaced on a regularly scheduled maintenance cycle. That eliminates the need the need to dispatch an IT administrator every time a component fails.
“We make a lot of use of automation tools, a self-healing network and zero-touch provisioning,” Bachar reports. “We’re continuously trying to improve on our operational expenses.”
Bachar says LinkedIn is willing to share its IT infrastructure expertise with anyone interested. In the meantime, the company’s next goal is to get ready to make the shift from 10G to 25G Ethernet server environments. Once that occurs, it’s only a matter of time before the network itself eventually moves up to 200G and beyond.
Of course, given the number of users LinkedIn is trying to support, making that move to 200/400G is likely to become a requirement a lot sooner than it will at the average IT organization.