a.k.a The morning Lori was wrong

I got an e-mail newsletter yesterday with a link to BEA's Virtualization TCO calculator. As my team is engaged in a lively debate regarding virtualization and its alleged benefits (you can tell which side of the fence I'm on at the moment) I visited the calculator to see what it would say. Then I sent the following results to the team, a smug smile on my face because the virtualized OS environment turned out to be more expensive than the non-virtualized environment.

 
TCO Summary
Non-Virtualized
Virtualized OS
Virtualized LVM
Server Hardware
$125,000
$125,000
$100,000
Software License/Support
$1,483,000
$2,058,313
$2,241,250
Software Administration
$270,000
$1,215,000
$378,000
Unplanned Downtime
$1,500,000
$375,000
$300,000
Data Center Real Estate
$27,977
$27,977
$18,651
Power and Cooling
$56,365
$56,365
$45,092
Total
$3,462,342
$3,857,655
$3,082,994
 

Someone, of course, pointed out that the power and cooling costs didn't quite jive. My thought was that maybe it does. After all, while you are reducing the number of servers in a virtualized architecture, you're also increasing the average utilization of those servers. That ought to even out in the end. But I didn't want to leave it to simply theory, I wanted some concrete proof.

So I did some research and found some real data1 with which I thought I could back up my theory, then plugged that data into a spreadsheet. I thought "Ha! I'll show you that this isn't the rosy picture so many virtualization supporters like to paint."

So here's where I get to admit I was wrong. Yeah, I know - hard to believe, but it's true. And here's the data behind that startling epiphany. Each grouping of data is based on consolidation of 10 servers with average CPU utilization of 40%, 65%, and 80% respectively. The resulting number of servers is the number required to maintain the same load overall, but at 95% utilization per server.

Number of Servers Amps per server Watts per server BTU/Hour per server Total Amps Total Watts Total BTU 40% AVG -> 95% AVG
10 1.5 325 1107 15 3250 11070 NON VIRTUALIZED
4 1.9 415 1414 7.6 1660 5656 VIRTUALIZED
Number of Servers Amps per server Watts per server BTU/Hour per server Total Amps Total Watts Total BTU 65% AVG -> 95% AVG
10 1.7 366 1249 17 3660 12490 NON VIRTUALIZED
7 1.9 415 1414 13.3 2905 9898 VIRTUALIZED
Number of Servers Amps per server Watts per server BTU/Hour per server Total Amps Total Watts Total BTU 80% AVG -> 95% AVG
10 1.8 391 1107 18 3910 11070 NON VIRTUALIZED
8 1.9 415 1414 15.2 3320 11312 VIRTUALIZED

It appears, then, that although there is a diminishing return on the amount of energy saved by consolidating physical servers and virtualizing the environment, still there is a return. Fewer servers == less power, even given that modern servers draw variable power based on utilization. Turns out that OS virtualization may indeed be more energy efficient in the long run.

Hmmmph. Well, there's still the issue of increased licensing and administrative costs.

The debate, I'm sure, will rage on.

1 Power consumption data based on HP DL360 G5 servers using HP's Power Calculator

Imbibing: Coffee