Thursday, July 17, 2008

Data Center Grade Energy Efficiency for the Rest of Us

James Hamilton has an interesting set of slides on his blog focusing on ways of improving efficiency of power consumption in large data centers. He examines factors end-to-end, from utility power distribution all the way down to server hardware itself. Data center size, geographic distribution, load characteristics, hardware profiles, power redundancy, etc. are all important variables to consider in the large data center space. An incredibly fascinating read.

What’s also interesting about what’s going on in the data center space is that it reflects in many ways a more laser-like focus on energy efficiency than one would typically encounter elsewhere. Given the scale that data center folks operate on, they’re thinking about the problems that many other types of organizations haven’t even dreamt of (yet). There’s a lot we can learn from what’s going on in that space, and perhaps even a few things large-, medium- and small-organizations can do to help increase their energy efficiency.

Of course, one thing I’ve always thought made sense was to “carpool” some types of server applications onto services like Amazon Web Services or Google Application Engine that are often served in-house (I’m not specifically endorsing any of these services – just throwing out a few possibilities). From an ecological standpoint, all of the energy (no pun intended) that folks who run data centers put into reducing power and cooling costs really does result in a significantly more efficient operation. Running your own application code in someone else’s data center (“carpooling” with many other applications that have been virtualized or are otherwise able to run well-isolated, securely and in parallel) means less energy wasted in your own data center (or more often, server room) in terms of power, cooling, and idle CPU cycles. This form of computing really is coming into fruition as a bona fide utility.

Naturally, not every application or service can be “carpooled” but given the excellent and in-depth work that’s been done in the data center, why replicate or re-invent the wheel? When planning or building new applications, in today’s world, it may make sense from both an ecological and economical standpoint (think of the reduced capital and operating expenses) to consider “carpooling” applications in an energy-efficient data center.

No comments: