Page 62 - How Cloud Computing Is Transforming Business and Why You Cant Afford to Be Left Behind

MANA GEMENT STRATEGIES F O R THE CL OUD R EV OL UTION

Cloud data centers have overcome the barriers to build-
ing large clusters that continue to scale upward in processing
power as more servers are added. They may lose a small por-
tion of that power; perhaps 10 or 15 percent of each server’s
processing has to be sacrificed to overhead. But a great deal
of the power of each added server gets added to the collective

power of the cluster. A lot of design skill has gone into allow-
ing the cluster to scale out as the number of nodes grows into
the thousands. Google, Amazon.com, Microsoft, IBM, Rack-
space, and other suppliers appear to have mastered the art of
building clusters for public cloud computing purposes.
Writing in 1995, Pfister with foresight concluded: “Sud-
denly, large but practical-sized agglomerations of microproces-

sors didn’t just equal big machine performance or provide it
more cheaply. They clearly became the way to exceed even the
biggest and superest computer speeds by large amounts.”
There are many people in Silicon Valley who would be
building cloud computers—small clusters—in their base-
ments at night, if only it were easy to do. All too soon the
would-be cluster builder reaches the territory where hic sunt
dracones, and the performance of his cluster is consumed by
its own design.

If today’s cloud suppliers have avoided this pitfall, how big
are their clusters? Yahoo! runs one type of cloud software,
called Hadoop, a data analysis system, on clusters of 4,000
computers. But the size of a single cluster is no longer the
point. The cloud supplier of the future will build a series of
data centers around the world, putting compute power close

42