Facebook details custom data centre risks

Facebook spent $US606 million on its data centres in 2011 and expects that to rise to at least $US1.6 billion this year as it rolls out more of its highly-customised data hubs. 

The social network's 2011 capital expenditure included the purchase of servers, networking equipment, storage infrastructure and the construction of data centres, it revealed in its first Securities and Exchange Commission filing yesterday. 

Facebook also locked in $US500 million under contractual committments related to data centre equipment as of December 31, 2011, bringing its total spend on data centres in 2011 to just over $US1.1 billion.

The filing highlighted how critical data centre infrastructure is to Facebook's future, with technical and cost risks posing a potential threat to its profitability.

The value of its installed network equipment rose from $US478 million in 2010 to just over $US1 billion in 2011, while its expenditure on "construction in progress" last year nearly doubled to $US327 million.

New construction projects included its US data centres in Oregon and North Carolina, its first non-US data centre underway in the Northern Swedish town of Luleå and its new headquarters in Menlo Park, California. 

Facebook warned investors that it expected to "incur increasing costs, in particular for servers, storage, power and data centers" to support its future growth, which may impact profitability.   

In April 2011 Facebook began serving products from its own data centre in Prineville, Oregon with "servers specifically designed for us".

Facebook is one of several major internet companies that are bypassing traditional hardware vendors for original development manufacturers, primarily in Taiwan.

Facebook will continue down this track with data centres it designs and owns but warned of unforeseen difficulties due to its "limited experience in this area". 

"The infrastructure expansion we are undertaking is complex, and unanticipated delays in the completion of these projects or availability of components may lead to increased project costs, operational inefficiencies, or interruptions in the delivery or degradation of the quality of our products."

On the flip side, its optimised data centre facilities, servers and software delivered "significant" energy savings and cut its server operation costs compared with traditional servers and leased data centres.

Facebook did this by removing non-essential components from servers to reduce cost and improve serviceability, improving server cooing and power distribution to minimise power loss in servers and its data centres, and optimising its power distribution system.