Uptime: Data centers are hungrier than ever for power
In a recent podcast, Uptime Institute founder and executive director Kenneth Brill talked about the dire power situation today’s data center industry is in. Specifically, Brill said that Uptime research has found that the top third of all data center sites it looked at (meaning about 50 sites that averaged 60,000 square feet in size) had a 24% compound annual growth rate in power consumption in 2006 and 2007. That’s a huge jump from the 7.5% energy growth rate from 1999 to 2005 for those same data centers.
Why does this matter? In the report on data center energy consumption that the EPA sent to Congress in August, it reported that the energy growth rate in data centers would be 15.7% from 2006 to 2010, an increase it estimated would require the building of 10 1,000-megawatt power plants, which are nothing to sneeze at in terms of acquiring the permits and actually building the facilities. This was data that Brill said he previously ignored this data as “somewhat alarmist.” But with Uptime’s recent research of its own sites, he’s no longer doing that.
“I can absolutely tell you that the top third of data centers in the Site Uptime Network in 2006 and 2007 is experiencing a 24% compound annual compound growth rate in their power consumption, which is unprecedented,” he said.
Brill estimates that 30 more 1,000-megawatt power plants will need to be built by 2015 just to support this IT energy growth. Brill continued in his talk to say that there is no “technology silver bullet” that will make building data center facilities cheaper when looking at its power capacity compared to price. What he instead suggests is a “tuning up” of what users already have.
“There are opportunities for efficiency improvements all over, but it requires tuning up things and that is something that we as human beings seem to be constitutionally unable to do,” he said. “We seem to prefer buying new stuff instead of tuning up what we already have, and yet tuning up what we already have is the most profitable thing we could do.”
Posted: March 20th, 2008 under Data center physical infrastructure, Data Center Power, Data center power efficiency.
Kenneth Brill states that there is “no technology silver bullet” to solve this energy problem, and I agree. Keep in mind that data is growing at almost exponential rates and this data requires spinning disks to sit on. Studies show that for every $1 consumed in power for servers and storage, an additional $2.33 will be spent on UPS, chillers, humidifiers, etc. We spend more than twice as much in power to “manage” the data center than to manage the data. That simply means that any reduction in the power to the drives and servers will have a dramatic impact on the entire power consumed in the data center.
A very common sense approach is to virtualize - not just servers, but storage as well. One storage manufacturer has got this down cold - Compellent. Let me explain why: traditional storage “allocates” space on the disk as a LUN. This “cannister” then is vacant, with nothing in it, until data is poured into it. As a result, most disk space that is eaten up in a data center is “allocated” space, not space with data on it. Only two companies address this with real Thin Provisioning - 3PAR and Compellent - all the others claim Thin Provisioning, but thy dont even come close to solving the problem like these two innovators have. Real Thin Provisioning, takes the space from the disk pool when the data is written to it, not at LUN creation. This simply saves dramatic amounts of disk space over all the other “Traditional” solutions.
Now add to it Compellent’s Automated Tiered ILM - this one-of-a-kind feature moves unused blocks automatically from expensive Fibre Drives to low cost SATA drives automatically. And I am talking about blocks, not files, records, or volumes. Blocks. In this type of solution you can have the header of an email (say 2k in size) sitting on high speed 15K spindles and the attached 1MB Powerpoint is on low cost SATA. This is done without intervention - and it works. So, instead of buying hundreds of TB’s of Fibre drives, you buy 10-20TB of Fibre Drives and purchase 200TB of slower spinning, bigger capacity SATA drives.
By combining the two “Thin Provisioning” and Automated Tiered ILM - these guys reduce the source of the power consumption and bring a dramatic impact to this ever growing problem. Hat’s off to the innovators !!!
Paul Clifford
Davenport Group
www.davenportgroup.com
Comment by Paul Clifford — March 21, 2008 @ 11:53 am
Paul
That’s an inefficient data centre, the metrics that I have are about 1 to 1 (ie 1 W of electricity to remove 1 W of heat used to power the machines). A good data centre should be nearer 60:40. If you have a high power input density, there is a trap that you may need a spare cooling system as if the primary fails, the equipment will melt
I like the automatic classification of data - I’ve never seen any manual tagging of data with importance work, except in government agencies.
If you want to use virtualisation (servers or storage) to save any resource, you’ve got to account for any increase in demand from a cheaper resource. For example, I’ve seen server virtualisation being used by DBAs as a quick / cheaper way of dumping an instance of a database. It is important to ensure that your cross charging represents actual costs, otherwise the users will break your best laid plans.
Tim
Comment by Tim Coote — March 26, 2008 @ 9:34 am