<s>
A	O
server	B-Operating_System
farm	I-Operating_System
or	O
server	B-Architecture
cluster	I-Architecture
is	O
a	O
collection	O
of	O
computer	B-Application
servers	I-Application
,	O
usually	O
maintained	O
by	O
an	O
organization	O
to	O
supply	O
server	B-Application
functionality	O
far	O
beyond	O
the	O
capability	O
of	O
a	O
single	O
machine	O
.	O
</s>
<s>
At	O
the	O
optimum	O
performance	O
level	O
,	O
a	O
server	B-Operating_System
farm	I-Operating_System
has	O
enormous	O
financial	O
and	O
environmental	O
costs	O
.	O
</s>
<s>
They	O
often	O
include	O
backup	O
servers	O
that	O
can	O
take	O
over	O
the	O
functions	O
of	O
primary	B-Operating_System
servers	I-Operating_System
that	O
may	O
fail	O
.	O
</s>
<s>
Server	B-Operating_System
farms	I-Operating_System
are	O
typically	O
collocated	O
with	O
the	O
network	B-Protocol
switches	I-Protocol
and/or	O
routers	B-Protocol
that	O
enable	O
communication	O
between	O
different	O
parts	O
of	O
the	O
cluster	O
and	O
the	O
cluster	O
's	O
users	O
.	O
</s>
<s>
Server	B-Application
"	O
farmers	O
"	O
typically	O
mount	O
computers	O
,	O
routers	B-Protocol
,	O
power	O
supplies	O
and	O
related	O
electronics	O
on	O
19-inch	B-Application
racks	I-Application
in	O
a	O
server	B-Application
room	I-Application
or	O
data	B-Operating_System
center	I-Operating_System
.	O
</s>
<s>
Server	B-Operating_System
farms	I-Operating_System
are	O
commonly	O
used	O
for	O
cluster	B-Architecture
computing	I-Architecture
.	O
</s>
<s>
Many	O
modern	O
supercomputers	B-Architecture
comprise	O
giant	O
server	B-Operating_System
farms	I-Operating_System
of	O
high-speed	O
processors	O
connected	O
by	O
either	O
Gigabit	O
Ethernet	O
or	O
custom	O
interconnects	O
such	O
as	O
Infiniband	B-Architecture
or	O
Myrinet	B-General_Concept
.	O
</s>
<s>
Web	O
hosting	O
is	O
a	O
common	O
use	O
of	O
a	O
server	B-Operating_System
farm	I-Operating_System
;	O
such	O
a	O
system	O
is	O
sometimes	O
collectively	O
referred	O
to	O
as	O
a	O
web	B-Operating_System
farm	I-Operating_System
.	O
</s>
<s>
Other	O
uses	O
of	O
server	B-Operating_System
farms	I-Operating_System
include	O
scientific	O
simulations	O
(	O
such	O
as	O
computational	O
fluid	O
dynamics	O
)	O
and	O
the	O
rendering	O
of	O
3D	O
computer	O
generated	O
imagery	O
(	O
see	O
render	B-Operating_System
farm	I-Operating_System
)	O
.	O
</s>
<s>
Server	B-Operating_System
farms	I-Operating_System
are	O
increasingly	O
being	O
used	O
instead	O
of	O
or	O
in	O
addition	O
to	O
mainframe	B-Architecture
computers	I-Architecture
by	O
large	O
enterprises	O
.	O
</s>
<s>
In	O
large	O
server	B-Operating_System
farms	I-Operating_System
,	O
the	O
failure	O
of	O
an	O
individual	O
machine	O
is	O
a	O
commonplace	O
event	O
:	O
large	O
server	B-Operating_System
farms	I-Operating_System
provide	O
redundancy	O
,	O
automatic	O
failover	B-General_Concept
,	O
and	O
rapid	O
reconfiguration	O
of	O
the	O
server	B-Architecture
cluster	I-Architecture
.	O
</s>
<s>
The	O
performance	O
of	O
the	O
largest	O
server	B-Operating_System
farms	I-Operating_System
(	O
thousands	O
of	O
processors	O
and	O
up	O
)	O
is	O
typically	O
limited	O
by	O
the	O
performance	O
of	O
the	O
data	B-Operating_System
center	I-Operating_System
's	O
cooling	B-General_Concept
systems	O
and	O
the	O
total	O
electricity	O
cost	O
rather	O
than	O
by	O
the	O
processors	O
 '	O
performance	O
.	O
</s>
<s>
Computers	O
in	O
server	B-Operating_System
farms	I-Operating_System
run	O
24/7	O
and	O
consume	O
large	O
amounts	O
of	O
electricity	O
.	O
</s>
<s>
Also	O
,	O
for	O
high	O
availability	O
systems	O
that	O
must	O
run	O
24/7	O
(	O
unlike	O
supercomputers	B-Architecture
that	O
can	O
be	O
power-cycled	O
to	O
demand	O
,	O
and	O
also	O
tend	O
to	O
run	O
at	O
much	O
higher	O
utilizations	O
)	O
,	O
there	O
is	O
more	O
attention	O
to	O
power-saving	O
features	O
such	O
as	O
variable	O
clock-speed	O
and	O
the	O
ability	O
to	O
turn	O
off	O
both	O
computer	O
parts	O
,	O
processor	O
parts	O
,	O
and	O
entire	O
computers	O
(	O
WoL	B-Protocol
and	O
virtualization	B-General_Concept
)	O
according	O
to	O
demand	O
without	O
bringing	O
down	O
services	O
.	O
</s>
<s>
The	O
network	O
connecting	O
the	O
servers	O
in	O
a	O
server	B-Operating_System
farm	I-Operating_System
is	O
also	O
an	O
essential	O
factor	O
in	O
overall	O
performance	O
,	O
especially	O
when	O
running	O
applications	O
that	O
process	O
massive	O
volumes	O
of	O
data	O
.	O
</s>
<s>
The	O
EEMBC	O
EnergyBench	O
,	O
SPECpower	O
,	O
and	O
the	O
Transaction	B-General_Concept
Processing	I-General_Concept
Performance	I-General_Concept
Council	I-General_Concept
TPC-Energy	O
are	O
benchmarks	O
designed	O
to	O
predict	O
performance	O
per	O
watt	O
in	O
a	O
server	B-Operating_System
farm	I-Operating_System
.	O
</s>
<s>
Some	O
servers	O
include	O
power	O
tracking	O
hardware	O
so	O
the	O
people	O
running	O
the	O
server	B-Operating_System
farm	I-Operating_System
can	O
measure	O
the	O
power	O
used	O
by	O
each	O
server	B-Application
.	O
</s>
<s>
The	O
power	O
used	O
by	O
the	O
entire	O
server	B-Operating_System
farm	I-Operating_System
may	O
be	O
reported	O
in	O
terms	O
of	O
power	O
usage	O
effectiveness	O
or	O
data	B-Operating_System
center	I-Operating_System
infrastructure	O
efficiency	O
.	O
</s>
<s>
For	O
this	O
reason	O
,	O
the	O
siting	O
of	O
a	O
server	B-Operating_System
farm	I-Operating_System
can	O
be	O
as	O
important	O
as	O
processor	O
selection	O
in	O
achieving	O
power	O
efficiency	O
.	O
</s>
<s>
Iceland	O
,	O
which	O
has	O
a	O
cold	O
climate	O
all	O
year	O
as	O
well	O
as	O
cheap	O
and	O
carbon-neutral	O
geothermal	O
electricity	O
supply	O
,	O
is	O
building	O
its	O
first	O
major	O
server	B-Operating_System
farm	I-Operating_System
hosting	O
site	O
.	O
</s>
<s>
Other	O
countries	O
with	O
favorable	O
conditions	O
,	O
such	O
as	O
Canada	O
,	O
Finland	O
,	O
Sweden	O
and	O
Switzerland	O
,	O
are	O
trying	O
to	O
attract	O
cloud	O
computing	O
data	B-Operating_System
centers	I-Operating_System
.	O
</s>
