recouping our investment in the first
year,” Dr. Furlani says. “They came back
and said, ‘That’s true, but only if you
remove the old servers first.’”
His team countered the skepticism
with a promise to get rid of the antiquated, power-hungry machines.
The project calls for replacing 256 of
the center’s existing servers with 128
high-efficiency ones. The shift is expected
to cut energy costs while increasing the
center’s total compute capacity from 13
teraflops (trillion floating-point operations per second) to 20 teraflops—a
more than 50 percent jump.
Dr. Furlani originally expected to have
the servers installed in mid-2009, but the
project was delayed to wait for the release
of new processors that promised to significantly improve performance.
In the meantime, his team obtained
loaner servers to measure the power
consumption under various levels of use
to help ensure they would meet the
promised savings. The vendors offer
online configuration tools that predict
power-consumption levels, “but we just
wanted to be sure they were accurate.”
Dr. Furlani and his team installed a
loaner model in the center and measured
its power consumption at various levels of
use. They then compared those rates to the
results forecast by the online tool.
“This is especially critical because
oftentimes the model you test is not the
model you end up buying due to the rapid
rate at which new products are introduced
into the marketplace,” he explains. “We
found that indeed, the online tools were
accurate, and therefore had great confidence that the final installation will realize
the desired energy savings.”
To keep the facilities team connected,
Dr. Furlani’s team has made energy-consumption tracking and reporting an
ongoing process. The center developed a
website that will track in real time how
much energy it’s using and compares
that number to the amount it would
have been using with the old servers.
DEATH TO THE
Securing support for green IT
projects doesn’t end with the
people who control the cash.
When the IT department starts
merging servers, eliminating
applications and co-locating
systems, the people—from
marketing to payroll—who
rely on those technologies to
get their work done tend to
become very nervous.
When someone’s job depends on the availability and
performance of a certain application, there’s bound to
be some resistance if their dedicated server—with its
dedicated backup—is moved into a shared environment. Fair or not, the assumption is that they’ll have to
worry about limited access or higher rates of failure,
says Chris Mines of Forrester Research.
To alleviate concerns—which are usually unfounded—
he suggests making people aware of the true energy
cost of their own technology use. For example, he
recently worked with a CIO who brought the company’s
utility bill onto his own budget and then allocated the
energy costs for powering and cooling the servers to the
“Some internal customers saw a 50 percent jump
in their internal billing. All of a sudden, those server
huggers became server shunners and virtualization
was a no-brainer,” Mr. Mines says. “Saving energy
became a great incentive for server virtualization.”
Not every business has the flexibility to assign
electricity costs to department budgets, but simply
sharing the true cost of an application can be an eye-opener.
“When you have to pay for your own energy it
makes you a lot more mindful of your use,” says Mark
Bramfitt of PG&E.
The display also includes a running tabulation of how much money the university is saving from reduced power consumption. And once the servers are
installed, the data will also be showcased
on a flat-screen display at the center.
“We’ve got a great story to tell and we
have the metrics to tell it, so why would
we hide that in a report that not many
people would have access to,” Dr. Furlani
says. “The display, and especially the website, make people more aware that IT can
be an expensive business, and accordingly
we should look for cost savings whenever
For many companies feeling the pinch,
all those servers are becoming a big target.