8
continued to build out Prineville
and started construction on our
future sites in Forest City, North
Carolina; Luleå, Sweden; and
Altoona, Iowa. But in the same
spirit of hacking and iterating
that Facebook brings to its
product development, our teams
are always looking for ways to
carve out new efficiency gains.
One small example can be found
in how we evolved the
evaporative elements of our
cooling system. In our first
building in Prineville, we built a
wall of super-fine misters into
one of the stages of the air
handling process in the
penthouse. This system was
effective, but it required that the
water be treated through a
reverse-osmosis system to ensure
that the misting nozzles don’t
clog over time. For Building 2,
we replaced the wall of misters
with a wall of evaporative media
that we can soak with water
directly from the well we have on
site. Not only did this design
iteration allow us to remove the
RO system from future buildings,
but it also allowed us to lower
our water consumption by an
order of magnitude. Our water
usage efficiency (WUE) over the
last 12 months in Prineville is
.028 L/kWh.
In many ways, this example of
iterating from one data center
building to another illustrates the
ethos that pervades how we
approach data center design:
strive for simplicity and
efficiency; focus on speed and
flexibility; and be open and
collaborative. This approach is at
the core of everything we do,
and its impact is undeniable: Our
drive for greater data center
efficiency has played a big role in
helping Facebook save more
than $1.2 billion in infrastructure
costs over the last three years.
Simplicity Breeds
Efficiency
Steve Jobs once said, “Simple
can be harder than complex: You
have to work hard to get your
thinking clean to make it simple.
But it’s worth it in the end
because once you get there, you
can move mountains.” Simplicity
is at the core of our data center
design philosophy, in no small
part because it’s crucial to
efficiently managing the
hundreds of thousands of servers
in our fleet.
A peek into a Facebook data
center will show what hardware
we deploy, but it’s also important
to see what’s not there. From the
beginning, our hardware teams
have practiced a “vanity-free”
approach to design, including
only the features that are
absolutely required to the
function of the device. This
approach has led to a number of
“common sense” decisions, such
as removing the plastic bezels
from the front of our servers to
allow the fans at the back of the
servers to draw air across the
components without expending
as much energy.
It’s also led us to designs that are
more innovative. Our custom-
designed “Knox” storage server
is comprised of a storage disk
and a compute server from our
“Winterfell” web server. By
moving to our new “Honey
Badger” micro server module in
Knox, we eliminate need for
discrete compute servers. In
addition to reducing the power
draw and lowering cost on a per-
server basis, we are able to
increase our rack density from
nine to 15 servers per rack,
further improving our total cost
of ownership.
Removing a little bit of plastic or
opening up a little more room in
our racks might not seem like it
would make much of a
7X24 MAGAZINE FALL 2014
