reverse caching saves trees

Energy is a significant cost of running data centers. Over a three-year operating period at typical U.S. power costs, a server’s acquisition cost is about equal to its power cost.[1] One documented estimate puts the annual power cost of U.S. data centers at two billion dollars in 2003.[2] IT power costs should be incorporated into a sensible evaluation of IT budgets and considered in energy policy.

Reverse caching can significantly reduce data center energy costs. At the State of the Net Conference this past Wednesday, Dick Sullivan of EMC stated that 70% of data on high-performance drives in data centers hasn’t been touched in the past ninety days. Caching traditionally moves some currently relevant data to relatively fast memory. Reverse caching goes the other way. Moving data unlikely to be used to energy efficient storage (a kinetically idle or unplugged drive, or a dismounted tape) saves significant costs.

At the far end of reverse caching are major issues of digital preservation. Reverse caching puts digital preservation into a framework of shorter-run operation and maintenance issues. That may be a valuable management reform. Digital preservation is probably a larger business opportunity than reverse caching. Reverse caching might help to make digital preservation more prominent in (short-run) management strategy.

Update:

1) I fixed a few mistakes in the earlier version of this post.

2) Chuck’s Blog has a good discussion of data center power usage.

3) Jonathon Koomey, with AMD sponsorship, has recently estimated total server power consumption in the U.S. in 2005 as 45 billion kWh. That represents 1.2% of total U.S. electricity consumption, about the same amount of power that color televisions consume (Koomey (2007), p. i). At an electricity cost of $60 per MWh, that power costs $2.7 billion. Koomey’s server power estimate includes power for cooling and auxiliary equipment associated with servers. It doesn’t include data storage and network equipment power, which Koomey suggests accounts for 20-40% of data center power consumption (see p. 2). Koomey’s server power estimate also does not include power for custom-built servers. Google’s custom-built servers, if included, might increase the total power consumption figure by 1.7% (see p. 3).

Koomey, Jonathan G. (2007), Estimating Total Power Consumption by Servers in the U.S. and the World (pdf), Final Report, Feb. 15.

[1] The Real Story about Dynamic Smart Cooling, Fact 1, citing HP, Christopher Malone, PhD, Christian Belady, P.E., “Metrics to Characterize Data Center & IT Equipment Energy Use”, Digital Power Forum, Richardson, TX (September 2006).

[2] Jeffrey S. Chase, Darrell C. Anderson, Prachi N. Thakar, Amin M. Vahdat, Ronald P. Doyle, “Managing energy and server resources in hosting centers,” ACM Symposium on Operating Systems Principles, Proceedings of the eighteenth ACM symposium on Operating systems principles, Banff, Alberta, Canada, 2001, pp. 103 – 116, (available pdf). Citing Jennifer D. Mitchell-Jackson, Energy Needs in an Internet Economy: A Closer Look at Data Centers, Master’s thesis, Energy and Resources Group, University of California at Berkeley, July 2001. This estimate used a power cost of $100 per MWh. Where power supply costs are higher, the power consumption cost of a server would be proportionally greater.