Page 209 -

The Greening of IT
174 How Companies Can Make a Difference for the Environment

required for a particular compute capacity was reduced, effectively doubling
the space in the data center.
To justify the upgrade, the business case normalized the 170 kilowatt
power addition to 4,250 square feet, using a conversion factor of 40 watts per
square feet, the average density of the existing data center.
Because we did not create physical space with this additional power, we
had to replace servers with higher-density servers. We continued our direc-
tion to replace single CPU servers with multiple CPU servers. The dominant
new server was a 16-processor server that took about 8U of rack space.
Physical space was not created with the additional power. Existing servers
were replaced with higher-density servers. The direction was to replace single
CPU servers with multiple CPU servers. This improvement led to requiring
half of the space that the 1U servers required. The combination of additional
power and denser servers reclaimed space for additional equipment.

Step 3. Virtualize
The subject data center consisted of many different accounts. Each of these
accounts had architectural control over the type of hardware they used in ptg
their part of the data center. Over the years, we started working with the
accounts to right-size their servers during server-refresh periods, and to move
to a scale-up model (larger servers) so that we could leverage virtualization
techniques to increase server utilization.
In the beginning of 2006, a study was conducted on the 1,500 UNIX vari-
ant servers that were in the data center. The survey indicated that 59 percent
of the single CPU servers had a monthly utilization of less than 5 percent! The
majority of the servers were single-processor servers that had been sized for the
peak workload of the application. The following chart (see Figure 10.4)
provides further insight into the distribution of utilization for the one-way,
two-way, and four-way physical servers that were installed at that time.
In essence, what that meant was that on average, we had 20 times the
amount of distributed equipment on the floor than was needed. Imagine
finding a way to leverage those untapped resources and eliminate 80 percent
of your equipment!
Significant increases in the penetration of virtual technology began at this
site. This included installing 8-way and 16-way servers. By March 2007, the
percentage of servers with less than 5 percent monthly utilization had
dropped to less than 25 percent. By March 2008, this number had dropped
to less than 14 percent.