The Origins and Future of Distributed Computing and Clouds - The Practice of Cloud System Administration - page 460

Costs
If distributed computing could be more reliable and faster, could it be more cost effective,
too? The cost of one highly reliable and expensive machine compared very well to the cost
of enough commodity machines to provide equivalent capacity. In fact, the cost per unit of
capacity was often 3 to 50 times less expensive using distributed computing. Remember
thatpreviouslywesaw50to66percentidlecapacitywiththelargeserverswhenusedin N
+ 1 and N + 2 configurations. Suppose an entire rack of machines was required to provide
the equivalent capacity. That would typically be 40 machines, with two held as spares. ( N
+ 2 redundancy). The “waste” now drops from 66 percent to 5 percent. That shrinking of
idle capacity gives the distributed computing design a head start. Factoring in the power of
a commodity market to drive down the cost of components improves the situation further.
Moreover, one gets a volume discount when purchasing many computers—something you
can't do when buying one or two large machines.
Inthe2003article“WebSearchforaPlanet:TheGoogleClusterArchitecture,”Barroso
et al. wrote:
The cost advantages of using inexpensive, PC-based clusters over high-end multipro-
cessor servers can be quite substantial, at least for a highly parallelizable application
like ours. The example $278,000 rack contains 176 2-GHz Xeon CPUs, 176 Gbytes
of RAM, and 7 Tbytes of disk space. In comparison, a typical x86-based server con-
tains eight 2-GHz Xeon CPUs, 64 Gbytes of RAM, and 8 Tbytes of disk space; it
costs about $758,000. In other words, the multiprocessor server is about three times
more expensive but has 22 times fewer CPUs, three times less RAM, and slightly
more disk space. ( Barroso, Dean & Hölzle 2003 )
Further cost reductions came by stripping the machines down to the exact components
needed. These machines did not need video cards, audio cards, speakers, keyboards, USB
ports, or a fancy plastic bezel with a cool logo on the front. Even if eliminating an item
saved just a few dollars, when buying thousands of machines it added up. Some compan-
ies (Yahoo!) worked with vendors to build custom computers to their exact specifications,
while others grew large enough to design their own computers from scratch (Google).
All these changes altered the economics of computing. In fact, they are what enabled
the second web era.
Undertheoldeconomics,thelargerthescale,themoredisproportionatelyexpensivethe
system became. In the new economics, the larger the scale, the more opportunity for the
economicstoimprove.Insteadofsuper-linearcostcurves,costgrowthwasclosertolinear.