<s>
A	O
computer	B-Architecture
cluster	I-Architecture
is	O
a	O
set	O
of	O
computers	O
that	O
work	O
together	O
so	O
that	O
they	O
can	O
be	O
viewed	O
as	O
a	O
single	O
system	O
.	O
</s>
<s>
Unlike	O
grid	O
computers	O
,	O
computer	B-Architecture
clusters	I-Architecture
have	O
each	O
node	B-Protocol
set	O
to	O
perform	O
the	O
same	O
task	O
,	O
controlled	O
and	O
scheduled	O
by	O
software	O
.	O
</s>
<s>
The	O
components	O
of	O
a	O
cluster	O
are	O
usually	O
connected	O
to	O
each	O
other	O
through	O
fast	O
local	B-General_Concept
area	I-General_Concept
networks	I-General_Concept
,	O
with	O
each	O
node	B-Protocol
(	O
computer	O
used	O
as	O
a	O
server	O
)	O
running	O
its	O
own	O
instance	O
of	O
an	O
operating	B-General_Concept
system	I-General_Concept
.	O
</s>
<s>
In	O
most	O
circumstances	O
,	O
all	O
of	O
the	O
nodes	B-Protocol
use	O
the	O
same	O
hardware	O
and	O
the	O
same	O
operating	B-General_Concept
system	I-General_Concept
,	O
although	O
in	O
some	O
setups	O
(	O
e.g.	O
</s>
<s>
using	O
Open	B-Operating_System
Source	I-Operating_System
Cluster	I-Operating_System
Application	I-Operating_System
Resources	I-Operating_System
(	O
OSCAR	O
)	O
)	O
,	O
different	O
operating	B-General_Concept
systems	I-General_Concept
can	O
be	O
used	O
on	O
each	O
computer	O
,	O
or	O
different	O
hardware	O
.	O
</s>
<s>
Computer	B-Architecture
clusters	I-Architecture
emerged	O
as	O
a	O
result	O
of	O
convergence	O
of	O
a	O
number	O
of	O
computing	O
trends	O
including	O
the	O
availability	O
of	O
low-cost	O
microprocessors	O
,	O
high-speed	O
networks	O
,	O
and	O
software	O
for	O
high-performance	O
distributed	B-Architecture
computing	I-Architecture
.	O
</s>
<s>
They	O
have	O
a	O
wide	O
range	O
of	O
applicability	O
and	O
deployment	O
,	O
ranging	O
from	O
small	O
business	O
clusters	O
with	O
a	O
handful	O
of	O
nodes	B-Protocol
to	O
some	O
of	O
the	O
fastest	B-Operating_System
supercomputers	I-Operating_System
in	O
the	O
world	O
such	O
as	O
IBM	B-Device
's	I-Device
Sequoia	I-Device
.	O
</s>
<s>
Prior	O
to	O
the	O
advent	O
of	O
clusters	O
,	O
single	O
unit	O
fault	B-General_Concept
tolerant	I-General_Concept
mainframes	B-Architecture
with	O
modular	B-Error_Name
redundancy	I-Error_Name
were	O
employed	O
;	O
but	O
the	O
lower	O
upfront	O
cost	O
of	O
clusters	O
,	O
and	O
increased	O
speed	O
of	O
network	O
fabric	O
has	O
favoured	O
the	O
adoption	O
of	O
clusters	O
.	O
</s>
<s>
In	O
contrast	O
to	O
high-reliability	O
mainframes	B-Architecture
clusters	O
are	O
cheaper	O
to	O
scale	O
out	O
,	O
but	O
also	O
have	O
increased	O
complexity	O
in	O
error	O
handling	O
,	O
as	O
in	O
clusters	O
error	O
modes	O
are	O
not	O
opaque	O
to	O
running	O
programs	O
.	O
</s>
<s>
The	O
computer	B-Architecture
clustering	I-Architecture
approach	O
usually	O
(	O
but	O
not	O
always	O
)	O
connects	O
a	O
number	O
of	O
readily	O
available	O
computing	O
nodes	B-Protocol
(	O
e.g.	O
</s>
<s>
personal	O
computers	O
used	O
as	O
servers	O
)	O
via	O
a	O
fast	O
local	B-General_Concept
area	I-General_Concept
network	I-General_Concept
.	O
</s>
<s>
The	O
activities	O
of	O
the	O
computing	O
nodes	B-Protocol
are	O
orchestrated	O
by	O
"	O
clustering	O
middleware	O
"	O
,	O
a	O
software	O
layer	O
that	O
sits	O
atop	O
the	O
nodes	B-Protocol
and	O
allows	O
the	O
users	O
to	O
treat	O
the	O
cluster	O
as	O
by	O
and	O
large	O
one	O
cohesive	O
computing	O
unit	O
,	O
e.g.	O
</s>
<s>
via	O
a	O
single	B-Operating_System
system	I-Operating_System
image	I-Operating_System
concept	O
.	O
</s>
<s>
Computer	B-Architecture
clustering	I-Architecture
relies	O
on	O
a	O
centralized	O
management	O
approach	O
which	O
makes	O
the	O
nodes	B-Protocol
available	O
as	O
orchestrated	O
shared	O
servers	O
.	O
</s>
<s>
It	O
is	O
distinct	O
from	O
other	O
approaches	O
such	O
as	O
peer-to-peer	O
or	O
grid	B-Architecture
computing	I-Architecture
which	O
also	O
use	O
many	O
nodes	B-Protocol
,	O
but	O
with	O
a	O
far	O
more	O
distributed	B-Architecture
nature	I-Architecture
.	O
</s>
<s>
A	O
computer	B-Architecture
cluster	I-Architecture
may	O
be	O
a	O
simple	O
two-node	O
system	O
which	O
just	O
connects	O
two	O
personal	O
computers	O
,	O
or	O
may	O
be	O
a	O
very	O
fast	O
supercomputer	B-Architecture
.	O
</s>
<s>
A	O
basic	O
approach	O
to	O
building	O
a	O
cluster	O
is	O
that	O
of	O
a	O
Beowulf	B-Operating_System
cluster	I-Operating_System
which	O
may	O
be	O
built	O
with	O
a	O
few	O
personal	O
computers	O
to	O
produce	O
a	O
cost-effective	O
alternative	O
to	O
traditional	O
high-performance	B-Architecture
computing	I-Architecture
.	O
</s>
<s>
An	O
early	O
project	O
that	O
showed	O
the	O
viability	O
of	O
the	O
concept	O
was	O
the	O
133-node	O
Stone	B-Operating_System
Soupercomputer	I-Operating_System
.	O
</s>
<s>
The	O
developers	O
used	O
Linux	B-Application
,	O
the	O
Parallel	B-Operating_System
Virtual	I-Operating_System
Machine	I-Operating_System
toolkit	O
and	O
the	O
Message	B-Application
Passing	I-Application
Interface	I-Application
library	O
to	O
achieve	O
high	O
performance	O
at	O
a	O
relatively	O
low	O
cost	O
.	O
</s>
<s>
The	O
TOP500	B-Operating_System
organization	O
's	O
semiannual	O
list	O
of	O
the	O
500	O
fastest	B-Operating_System
supercomputers	I-Operating_System
often	O
includes	O
many	O
clusters	O
,	O
e.g.	O
</s>
<s>
the	O
world	O
's	O
fastest	O
machine	O
in	O
2011	O
was	O
the	O
K	B-Device
computer	I-Device
which	O
has	O
a	O
distributed	B-Operating_System
memory	I-Operating_System
,	O
cluster	O
architecture	O
.	O
</s>
<s>
The	O
formal	O
engineering	O
basis	O
of	O
cluster	B-Architecture
computing	I-Architecture
as	O
a	O
means	O
of	O
doing	O
parallel	O
work	O
of	O
any	O
sort	O
was	O
arguably	O
invented	O
by	O
Gene	O
Amdahl	O
of	O
IBM	O
,	O
who	O
in	O
1967	O
published	O
what	O
has	O
come	O
to	O
be	O
regarded	O
as	O
the	O
seminal	O
paper	O
on	O
parallel	O
processing	O
:	O
Amdahl	B-Operating_System
's	I-Operating_System
Law	I-Operating_System
.	O
</s>
<s>
The	O
history	O
of	O
early	O
computer	B-Architecture
clusters	I-Architecture
is	O
more	O
or	O
less	O
directly	O
tied	O
into	O
the	O
history	O
of	O
early	O
networks	O
,	O
as	O
one	O
of	O
the	O
primary	O
motivations	O
for	O
the	O
development	O
of	O
a	O
network	O
was	O
to	O
link	O
computing	O
resources	O
,	O
creating	O
a	O
de	O
facto	O
computer	B-Architecture
cluster	I-Architecture
.	O
</s>
<s>
The	O
first	O
commercial	O
loosely	O
coupled	O
clustering	O
product	O
was	O
Datapoint	O
Corporation	O
's	O
"	O
Attached	O
Resource	O
Computer	O
"	O
(	O
ARC	O
)	O
system	O
,	O
developed	O
in	O
1977	O
,	O
and	O
using	O
ARCnet	B-Protocol
as	O
the	O
cluster	O
interface	O
.	O
</s>
<s>
Clustering	O
per	O
se	O
did	O
not	O
really	O
take	O
off	O
until	O
Digital	O
Equipment	O
Corporation	O
released	O
their	O
VAXcluster	B-Operating_System
product	O
in	O
1984	O
for	O
the	O
VMS	B-Operating_System
operating	I-Operating_System
system	I-Operating_System
.	O
</s>
<s>
The	O
ARC	O
and	O
VAXcluster	B-Operating_System
products	O
not	O
only	O
supported	O
parallel	O
computing	O
,	O
but	O
also	O
shared	O
file	B-Application
systems	I-Application
and	O
peripheral	O
devices	O
.	O
</s>
<s>
Two	O
other	O
noteworthy	O
early	O
commercial	O
clusters	O
were	O
the	O
Tandem	O
NonStop	O
(	O
a	O
1976	O
high-availability	B-General_Concept
commercial	O
product	O
)	O
and	O
the	O
IBM	O
S/390	O
Parallel	O
Sysplex	O
(	O
circa	O
1994	O
,	O
primarily	O
for	O
business	O
use	O
)	O
.	O
</s>
<s>
Within	O
the	O
same	O
time	O
frame	O
,	O
while	O
computer	B-Architecture
clusters	I-Architecture
used	O
parallelism	O
outside	O
the	O
computer	O
on	O
a	O
commodity	O
network	O
,	O
supercomputers	B-Architecture
began	O
to	O
use	O
them	O
within	O
the	O
same	O
computer	O
.	O
</s>
<s>
Following	O
the	O
success	O
of	O
the	O
CDC	B-Device
6600	I-Device
in	O
1964	O
,	O
the	O
Cray	B-Device
1	I-Device
was	O
delivered	O
in	O
1976	O
,	O
and	O
introduced	O
internal	O
parallelism	O
via	O
vector	B-Operating_System
processing	I-Operating_System
.	O
</s>
<s>
While	O
early	O
supercomputers	B-Architecture
excluded	O
clusters	O
and	O
relied	O
on	O
shared	B-Operating_System
memory	I-Operating_System
,	O
in	O
time	O
some	O
of	O
the	O
fastest	B-Operating_System
supercomputers	I-Operating_System
(	O
e.g.	O
</s>
<s>
the	O
K	B-Device
computer	I-Device
)	O
relied	O
on	O
cluster	O
architectures	O
.	O
</s>
<s>
Computer	B-Architecture
clusters	I-Architecture
may	O
be	O
configured	O
for	O
different	O
purposes	O
ranging	O
from	O
general	O
purpose	O
business	O
needs	O
such	O
as	O
web-service	O
support	O
,	O
to	O
computation-intensive	O
scientific	O
calculations	O
.	O
</s>
<s>
In	O
either	O
case	O
,	O
the	O
cluster	O
may	O
use	O
a	O
high-availability	B-General_Concept
approach	O
.	O
</s>
<s>
Note	O
that	O
the	O
attributes	O
described	O
below	O
are	O
not	O
exclusive	O
and	O
a	O
"	O
computer	B-Architecture
cluster	I-Architecture
"	O
may	O
also	O
use	O
a	O
high-availability	B-General_Concept
approach	O
,	O
etc	O
.	O
</s>
<s>
"	O
Load-balancing	B-Application
"	O
clusters	O
are	O
configurations	O
in	O
which	O
cluster-nodes	O
share	O
computational	O
workload	O
to	O
provide	O
better	O
overall	O
performance	O
.	O
</s>
<s>
For	O
example	O
,	O
a	O
web	O
server	B-Architecture
cluster	I-Architecture
may	O
assign	O
different	O
queries	O
to	O
different	O
nodes	B-Protocol
,	O
so	O
the	O
overall	O
response	O
time	O
will	O
be	O
optimized	O
.	O
</s>
<s>
However	O
,	O
approaches	O
to	O
load-balancing	B-Application
may	O
significantly	O
differ	O
among	O
applications	O
,	O
e.g.	O
</s>
<s>
a	O
high-performance	O
cluster	O
used	O
for	O
scientific	O
computations	O
would	O
balance	O
load	O
with	O
different	O
algorithms	O
from	O
a	O
web-server	O
cluster	O
which	O
may	O
just	O
use	O
a	O
simple	O
round-robin	B-Algorithm
method	I-Algorithm
by	O
assigning	O
each	O
new	O
request	O
to	O
a	O
different	O
node	B-Protocol
.	O
</s>
<s>
Computer	B-Architecture
clusters	I-Architecture
are	O
used	O
for	O
computation-intensive	O
purposes	O
,	O
rather	O
than	O
handling	O
IO-oriented	B-General_Concept
operations	O
such	O
as	O
web	O
service	O
or	O
databases	O
.	O
</s>
<s>
For	O
instance	O
,	O
a	O
computer	B-Architecture
cluster	I-Architecture
might	O
support	O
computational	B-Application
simulations	I-Application
of	O
vehicle	O
crashes	O
or	O
weather	O
.	O
</s>
<s>
Very	O
tightly	O
coupled	O
computer	B-Architecture
clusters	I-Architecture
are	O
designed	O
for	O
work	O
that	O
may	O
approach	O
"	O
supercomputing	B-Architecture
"	O
.	O
</s>
<s>
"	O
High-availability	B-General_Concept
clusters	I-General_Concept
"	O
(	O
also	O
known	O
as	O
failover	B-General_Concept
clusters	I-General_Concept
,	O
or	O
HA	B-General_Concept
clusters	I-General_Concept
)	O
improve	O
the	O
availability	O
of	O
the	O
cluster	O
approach	O
.	O
</s>
<s>
They	O
operate	O
by	O
having	O
redundant	O
nodes	B-Protocol
,	O
which	O
are	O
then	O
used	O
to	O
provide	O
service	O
when	O
system	O
components	O
fail	O
.	O
</s>
<s>
HA	B-General_Concept
cluster	I-General_Concept
implementations	O
attempt	O
to	O
use	O
redundancy	O
of	O
cluster	O
components	O
to	O
eliminate	O
single	B-Architecture
points	I-Architecture
of	I-Architecture
failure	I-Architecture
.	O
</s>
<s>
There	O
are	O
commercial	O
implementations	O
of	O
High-Availability	B-General_Concept
clusters	I-General_Concept
for	O
many	O
operating	B-General_Concept
systems	I-General_Concept
.	O
</s>
<s>
The	O
Linux-HA	B-Operating_System
project	O
is	O
one	O
commonly	O
used	O
free	B-Application
software	I-Application
HA	O
package	O
for	O
the	O
Linux	B-Application
operating	I-Application
system	I-Application
.	O
</s>
<s>
Fault	B-General_Concept
tolerance	I-General_Concept
(	O
the	O
ability	O
for	O
a	O
system	O
to	O
continue	O
working	O
with	O
a	O
malfunctioning	O
node	B-Protocol
)	O
allows	O
for	O
scalability	O
,	O
and	O
in	O
high-performance	O
situations	O
,	O
low	O
frequency	O
of	O
maintenance	O
routines	O
,	O
resource	O
consolidation	O
(	O
e.g.	O
</s>
<s>
RAID	B-Architecture
)	O
,	O
and	O
centralized	O
management	O
.	O
</s>
<s>
In	O
terms	O
of	O
scalability	O
,	O
clusters	O
provide	O
this	O
in	O
their	O
ability	O
to	O
add	O
nodes	B-Protocol
horizontally	O
.	O
</s>
<s>
This	O
means	O
that	O
more	O
computers	O
may	O
be	O
added	O
to	O
the	O
cluster	O
,	O
to	O
improve	O
its	O
performance	O
,	O
redundancy	O
and	O
fault	B-General_Concept
tolerance	I-General_Concept
.	O
</s>
<s>
This	O
can	O
be	O
an	O
inexpensive	O
solution	O
for	O
a	O
higher	O
performing	O
cluster	O
compared	O
to	O
scaling	O
up	O
a	O
single	O
node	B-Protocol
in	O
the	O
cluster	O
.	O
</s>
<s>
This	O
property	O
of	O
computer	B-Architecture
clusters	I-Architecture
can	O
allow	O
for	O
larger	O
computational	O
loads	O
to	O
be	O
executed	O
by	O
a	O
larger	O
number	O
of	O
lower	O
performing	O
computers	O
.	O
</s>
<s>
When	O
adding	O
a	O
new	O
node	B-Protocol
to	O
a	O
cluster	O
,	O
reliability	O
increases	O
because	O
the	O
entire	O
cluster	O
does	O
not	O
need	O
to	O
be	O
taken	O
down	O
.	O
</s>
<s>
A	O
single	O
node	B-Protocol
can	O
be	O
taken	O
down	O
for	O
maintenance	O
,	O
while	O
the	O
rest	O
of	O
the	O
cluster	O
takes	O
on	O
the	O
load	O
of	O
that	O
individual	O
node	B-Protocol
.	O
</s>
<s>
If	O
you	O
have	O
a	O
large	O
number	O
of	O
computers	O
clustered	O
together	O
,	O
this	O
lends	O
itself	O
to	O
the	O
use	O
of	O
distributed	O
file	B-Application
systems	I-Application
and	O
RAID	B-Architecture
,	O
both	O
of	O
which	O
can	O
increase	O
the	O
reliability	O
and	O
speed	O
of	O
a	O
cluster	O
.	O
</s>
<s>
One	O
of	O
the	O
issues	O
in	O
designing	O
a	O
cluster	O
is	O
how	O
tightly	O
coupled	O
the	O
individual	O
nodes	B-Protocol
may	O
be	O
.	O
</s>
<s>
For	O
instance	O
,	O
a	O
single	O
computer	O
job	O
may	O
require	O
frequent	O
communication	O
among	O
nodes	B-Protocol
:	O
this	O
implies	O
that	O
the	O
cluster	O
shares	O
a	O
dedicated	O
network	O
,	O
is	O
densely	O
located	O
,	O
and	O
probably	O
has	O
homogeneous	O
nodes	B-Protocol
.	O
</s>
<s>
The	O
other	O
extreme	O
is	O
where	O
a	O
computer	O
job	O
uses	O
one	O
or	O
few	O
nodes	B-Protocol
,	O
and	O
needs	O
little	O
or	O
no	O
inter-node	O
communication	O
,	O
approaching	O
grid	B-Architecture
computing	I-Architecture
.	O
</s>
<s>
In	O
a	O
Beowulf	B-Operating_System
cluster	I-Operating_System
,	O
the	O
application	O
programs	O
never	O
see	O
the	O
computational	O
nodes	B-Protocol
(	O
also	O
called	O
slave	O
computers	O
)	O
but	O
only	O
interact	O
with	O
the	O
"	O
Master	O
"	O
which	O
is	O
a	O
specific	O
computer	O
handling	O
the	O
scheduling	O
and	O
management	O
of	O
the	O
slaves	O
.	O
</s>
<s>
In	O
a	O
typical	O
implementation	O
the	O
Master	O
has	O
two	O
network	O
interfaces	O
,	O
one	O
that	O
communicates	O
with	O
the	O
private	O
Beowulf	B-Operating_System
network	O
for	O
the	O
slaves	O
,	O
the	O
other	O
for	O
the	O
general	O
purpose	O
network	O
of	O
the	O
organization	O
.	O
</s>
<s>
The	O
slave	O
computers	O
typically	O
have	O
their	O
own	O
version	O
of	O
the	O
same	O
operating	B-General_Concept
system	I-General_Concept
,	O
and	O
local	O
memory	O
and	O
disk	O
space	O
.	O
</s>
<s>
A	O
special	O
purpose	O
144-node	O
DEGIMA	B-Device
cluster	I-Device
is	O
tuned	O
to	O
running	O
astrophysical	O
N-body	O
simulations	O
using	O
the	O
Multiple-Walk	O
parallel	O
treecode	O
,	O
rather	O
than	O
general	O
purpose	O
scientific	O
computations	O
.	O
</s>
<s>
Due	O
to	O
the	O
increasing	O
computing	O
power	O
of	O
each	O
generation	O
of	O
game	O
consoles	O
,	O
a	O
novel	O
use	O
has	O
emerged	O
where	O
they	O
are	O
repurposed	O
into	O
High-performance	B-Architecture
computing	I-Architecture
(	O
HPC	O
)	O
clusters	O
.	O
</s>
<s>
Some	O
examples	O
of	O
game	O
console	O
clusters	O
are	O
Sony	B-General_Concept
PlayStation	I-General_Concept
clusters	I-General_Concept
and	O
Microsoft	O
Xbox	B-Application
clusters	O
.	O
</s>
<s>
Another	O
example	O
of	O
consumer	O
game	O
product	O
is	O
the	O
Nvidia	B-Device
Tesla	I-Device
Personal	I-Device
Supercomputer	I-Device
workstation	O
,	O
which	O
uses	O
multiple	O
graphics	O
accelerator	O
processor	O
chips	O
.	O
</s>
<s>
The	O
use	O
of	O
graphics	O
cards	O
(	O
or	O
rather	O
their	O
GPU	O
's	O
)	O
to	O
do	O
calculations	O
for	O
grid	B-Architecture
computing	I-Architecture
is	O
vastly	O
more	O
economical	O
than	O
using	O
CPU	O
's	O
,	O
despite	O
being	O
less	O
precise	O
.	O
</s>
<s>
Computer	B-Architecture
clusters	I-Architecture
have	O
historically	O
run	O
on	O
separate	O
physical	O
computers	O
with	O
the	O
same	O
operating	B-General_Concept
system	I-General_Concept
.	O
</s>
<s>
With	O
the	O
advent	O
of	O
virtualization	B-General_Concept
,	O
the	O
cluster	O
nodes	B-Protocol
may	O
run	O
on	O
separate	O
physical	O
computers	O
with	O
different	O
operating	B-General_Concept
systems	I-General_Concept
which	O
are	O
painted	O
above	O
with	O
a	O
virtual	O
layer	O
to	O
look	O
similar	O
.	O
</s>
<s>
The	O
cluster	O
may	O
also	O
be	O
virtualized	B-General_Concept
on	O
various	O
configurations	O
as	O
maintenance	O
takes	O
place	O
;	O
an	O
example	O
implementation	O
is	O
Xen	B-Operating_System
as	O
the	O
virtualization	B-General_Concept
manager	O
with	O
Linux-HA	B-Operating_System
.	O
</s>
<s>
As	O
the	O
computer	B-Architecture
clusters	I-Architecture
were	O
appearing	O
during	O
the	O
1980s	O
,	O
so	O
were	O
supercomputers	B-Architecture
.	O
</s>
<s>
One	O
of	O
the	O
elements	O
that	O
distinguished	O
the	O
three	O
classes	O
at	O
that	O
time	O
was	O
that	O
the	O
early	O
supercomputers	B-Architecture
relied	O
on	O
shared	B-Operating_System
memory	I-Operating_System
.	O
</s>
<s>
To	O
date	O
clusters	O
do	O
not	O
typically	O
use	O
physically	O
shared	B-Operating_System
memory	I-Operating_System
,	O
while	O
many	O
supercomputer	B-Architecture
architectures	O
have	O
also	O
abandoned	O
it	O
.	O
</s>
<s>
However	O
,	O
the	O
use	O
of	O
a	O
clustered	B-Application
file	I-Application
system	I-Application
is	O
essential	O
in	O
modern	O
computer	B-Architecture
clusters	I-Architecture
.	O
</s>
<s>
Examples	O
include	O
the	O
IBM	B-Application
General	I-Application
Parallel	I-Application
File	I-Application
System	I-Application
,	O
Microsoft	O
's	O
Cluster	O
Shared	O
Volumes	O
or	O
the	O
Oracle	B-Application
Cluster	I-Application
File	I-Application
System	I-Application
.	O
</s>
<s>
Two	O
widely	O
used	O
approaches	O
for	O
communication	O
between	O
cluster	O
nodes	B-Protocol
are	O
MPI	O
(	O
Message	B-Application
Passing	I-Application
Interface	I-Application
)	O
and	O
PVM	B-Operating_System
(	O
Parallel	B-Operating_System
Virtual	I-Operating_System
Machine	I-Operating_System
)	O
.	O
</s>
<s>
PVM	B-Operating_System
was	O
developed	O
at	O
the	O
Oak	O
Ridge	O
National	O
Laboratory	O
around	O
1989	O
before	O
MPI	O
was	O
available	O
.	O
</s>
<s>
PVM	B-Operating_System
must	O
be	O
directly	O
installed	O
on	O
every	O
cluster	O
node	B-Protocol
and	O
provides	O
a	O
set	O
of	O
software	O
libraries	O
that	O
paint	O
the	O
node	B-Protocol
as	O
a	O
"	O
parallel	B-Operating_System
virtual	I-Operating_System
machine	I-Operating_System
"	O
.	O
</s>
<s>
PVM	B-Operating_System
provides	O
a	O
run-time	O
environment	O
for	O
message-passing	O
,	O
task	O
and	O
resource	O
management	O
,	O
and	O
fault	O
notification	O
.	O
</s>
<s>
PVM	B-Operating_System
can	O
be	O
used	O
by	O
user	O
programs	O
written	O
in	O
C	B-Language
,	O
C++	O
,	O
or	O
Fortran	B-Application
,	O
etc	O
.	O
</s>
<s>
MPI	O
implementations	O
typically	O
use	O
TCP/IP	B-Protocol
and	O
socket	O
connections	O
.	O
</s>
<s>
MPI	O
is	O
now	O
a	O
widely	O
available	O
communications	O
model	O
that	O
enables	O
parallel	O
programs	O
to	O
be	O
written	O
in	O
languages	O
such	O
as	O
C	B-Language
,	O
Fortran	B-Application
,	O
Python	B-Language
,	O
etc	O
.	O
</s>
<s>
Thus	O
,	O
unlike	O
PVM	B-Operating_System
which	O
provides	O
a	O
concrete	O
implementation	O
,	O
MPI	O
is	O
a	O
specification	O
which	O
has	O
been	O
implemented	O
in	O
systems	O
such	O
as	O
MPICH	B-Library
and	O
Open	B-Library
MPI	I-Library
.	O
</s>
<s>
One	O
of	O
the	O
challenges	O
in	O
the	O
use	O
of	O
a	O
computer	B-Architecture
cluster	I-Architecture
is	O
the	O
cost	O
of	O
administrating	O
it	O
which	O
can	O
at	O
times	O
be	O
as	O
high	O
as	O
the	O
cost	O
of	O
administrating	O
N	O
independent	O
machines	O
,	O
if	O
the	O
cluster	O
has	O
N	O
nodes	B-Protocol
.	O
</s>
<s>
In	O
some	O
cases	O
this	O
provides	O
an	O
advantage	O
to	O
shared	B-Operating_System
memory	I-Operating_System
architectures	I-Operating_System
with	O
lower	O
administration	O
costs	O
.	O
</s>
<s>
This	O
has	O
also	O
made	O
virtual	B-Architecture
machines	I-Architecture
popular	O
,	O
due	O
to	O
the	O
ease	O
of	O
administration	O
.	O
</s>
<s>
This	O
is	O
an	O
area	O
of	O
ongoing	O
research	O
;	O
algorithms	O
that	O
combine	O
and	O
extend	O
MapReduce	B-Operating_System
and	O
Hadoop	B-Application
have	O
been	O
proposed	O
and	O
studied	O
.	O
</s>
<s>
When	O
a	O
node	B-Protocol
in	O
a	O
cluster	O
fails	O
,	O
strategies	O
such	O
as	O
"	O
fencing	B-Operating_System
"	O
may	O
be	O
employed	O
to	O
keep	O
the	O
rest	O
of	O
the	O
system	O
operational	O
.	O
</s>
<s>
Fencing	B-Operating_System
is	O
the	O
process	O
of	O
isolating	O
a	O
node	B-Protocol
or	O
protecting	O
shared	O
resources	O
when	O
a	O
node	B-Protocol
appears	O
to	O
be	O
malfunctioning	O
.	O
</s>
<s>
There	O
are	O
two	O
classes	O
of	O
fencing	B-Operating_System
methods	O
;	O
one	O
disables	O
a	O
node	B-Protocol
itself	O
,	O
and	O
the	O
other	O
disallows	O
access	O
to	O
resources	O
such	O
as	O
shared	O
disks	O
.	O
</s>
<s>
The	O
STONITH	B-Operating_System
method	O
stands	O
for	O
"	O
Shoot	O
The	O
Other	O
Node	B-Protocol
In	O
The	O
Head	O
"	O
,	O
meaning	O
that	O
the	O
suspected	O
node	B-Protocol
is	O
disabled	O
or	O
powered	O
off	O
.	O
</s>
<s>
For	O
instance	O
,	O
power	O
fencing	B-Operating_System
uses	O
a	O
power	O
controller	O
to	O
turn	O
off	O
an	O
inoperable	O
node	B-Protocol
.	O
</s>
<s>
The	O
resources	O
fencing	B-Operating_System
approach	O
disallows	O
access	O
to	O
resources	O
without	O
powering	O
off	O
the	O
node	B-Protocol
.	O
</s>
<s>
This	O
may	O
include	O
persistent	B-Operating_System
reservation	I-Operating_System
fencing	B-Operating_System
via	O
the	O
SCSI3	O
,	O
fibre	B-Architecture
channel	I-Architecture
fencing	B-Operating_System
to	O
disable	O
the	O
fibre	B-Architecture
channel	I-Architecture
port	O
,	O
or	O
global	B-Application
network	I-Application
block	I-Application
device	I-Application
(	O
GNBD	B-Application
)	O
fencing	B-Operating_System
to	O
disable	O
access	O
to	O
the	O
GNBD	B-Application
server	O
.	O
</s>
<s>
Load	B-Application
balancing	I-Application
clusters	O
such	O
as	O
web	O
servers	O
use	O
cluster	O
architectures	O
to	O
support	O
a	O
large	O
number	O
of	O
users	O
and	O
typically	O
each	O
user	O
request	O
is	O
routed	O
to	O
a	O
specific	O
node	B-Protocol
,	O
achieving	O
task	B-Operating_System
parallelism	I-Operating_System
without	O
multi-node	O
cooperation	O
,	O
given	O
that	O
the	O
main	O
goal	O
of	O
the	O
system	O
is	O
providing	O
rapid	O
user	O
access	O
to	O
shared	B-Operating_System
data	I-Operating_System
.	O
</s>
<s>
However	O
,	O
"	O
computer	B-Architecture
clusters	I-Architecture
"	O
which	O
perform	O
complex	O
computations	O
for	O
a	O
small	O
number	O
of	O
users	O
need	O
to	O
take	O
advantage	O
of	O
the	O
parallel	O
processing	O
capabilities	O
of	O
the	O
cluster	O
and	O
partition	O
"	O
the	O
same	O
computation	O
"	O
among	O
several	O
nodes	B-Protocol
.	O
</s>
<s>
Automatic	O
parallelization	O
of	O
programs	O
remains	O
a	O
technical	O
challenge	O
,	O
but	O
parallel	B-Application
programming	I-Application
models	I-Application
can	O
be	O
used	O
to	O
effectuate	O
a	O
higher	O
degree	B-Operating_System
of	I-Operating_System
parallelism	I-Operating_System
via	O
the	O
simultaneous	O
execution	O
of	O
separate	O
portions	O
of	O
a	O
program	O
on	O
different	O
processors	O
.	O
</s>
<s>
Tools	O
such	O
as	O
TotalView	O
were	O
then	O
developed	O
to	O
debug	O
parallel	O
implementations	O
on	O
computer	B-Architecture
clusters	I-Architecture
which	O
use	O
Message	B-Application
Passing	I-Application
Interface	I-Application
(	O
MPI	O
)	O
or	O
Parallel	B-Operating_System
Virtual	I-Operating_System
Machine	I-Operating_System
(	O
PVM	B-Operating_System
)	O
for	O
message	O
passing	O
.	O
</s>
<s>
The	O
University	O
of	O
California	O
,	O
Berkeley	O
Network	B-Architecture
of	I-Architecture
Workstations	I-Architecture
(	O
NOW	O
)	O
system	O
gathers	O
cluster	O
data	O
and	O
stores	O
them	O
in	O
a	O
database	O
,	O
while	O
a	O
system	O
such	O
as	O
PARMON	O
,	O
developed	O
in	O
India	O
,	O
allows	O
visually	O
observing	O
and	O
managing	O
large	O
clusters	O
.	O
</s>
<s>
Application	B-General_Concept
checkpointing	I-General_Concept
can	O
be	O
used	O
to	O
restore	O
a	O
given	O
state	O
of	O
the	O
system	O
when	O
a	O
node	B-Protocol
fails	O
during	O
a	O
long	O
multi-node	O
computation	O
.	O
</s>
<s>
This	O
is	O
essential	O
in	O
large	O
clusters	O
,	O
given	O
that	O
as	O
the	O
number	O
of	O
nodes	B-Protocol
increases	O
,	O
so	O
does	O
the	O
likelihood	O
of	O
node	B-Protocol
failure	O
under	O
heavy	O
computational	O
loads	O
.	O
</s>
<s>
Checkpointing	B-General_Concept
can	O
restore	O
the	O
system	O
to	O
a	O
stable	O
state	O
so	O
that	O
processing	O
can	O
resume	O
without	O
needing	O
to	O
recompute	O
results	O
.	O
</s>
<s>
The	O
Linux	B-Application
world	O
supports	O
various	O
cluster	O
software	O
;	O
for	O
application	O
clustering	O
,	O
there	O
is	O
distcc	B-Device
,	O
and	O
MPICH	B-Library
.	O
</s>
<s>
Linux	B-Operating_System
Virtual	I-Operating_System
Server	I-Operating_System
,	O
Linux-HA	B-Operating_System
-	O
director-based	O
clusters	O
that	O
allow	O
incoming	O
requests	O
for	O
services	O
to	O
be	O
distributed	O
across	O
multiple	O
cluster	O
nodes	B-Protocol
.	O
</s>
<s>
MOSIX	B-Operating_System
,	O
LinuxPMI	B-Operating_System
,	O
Kerrighed	B-Operating_System
,	O
OpenSSI	B-Operating_System
are	O
full-blown	O
clusters	O
integrated	O
into	O
the	O
kernel	B-Operating_System
that	O
provide	O
for	O
automatic	O
process	O
migration	O
among	O
homogeneous	O
nodes	B-Protocol
.	O
</s>
<s>
OpenSSI	B-Operating_System
,	O
openMosix	B-Operating_System
and	O
Kerrighed	B-Operating_System
are	O
single-system	B-Operating_System
image	I-Operating_System
implementations	O
.	O
</s>
<s>
Microsoft	B-Application
Windows	I-Application
computer	B-Architecture
cluster	I-Architecture
Server	O
2003	O
based	O
on	O
the	O
Windows	B-Device
Server	I-Device
platform	O
provides	O
pieces	O
for	O
High-Performance	B-Architecture
Computing	I-Architecture
like	O
the	O
Job	O
Scheduler	O
,	O
MSMPI	O
library	O
and	O
management	O
tools	O
.	O
</s>
<s>
gLite	B-Operating_System
is	O
a	O
set	O
of	O
middleware	O
technologies	O
created	O
by	O
the	O
Enabling	B-Operating_System
Grids	I-Operating_System
for	I-Operating_System
E-sciencE	I-Operating_System
(	O
EGEE	O
)	O
project	O
.	O
</s>
<s>
slurm	B-Operating_System
is	O
also	O
used	O
to	O
schedule	O
and	O
manage	O
some	O
of	O
the	O
largest	O
supercomputer	B-Architecture
clusters	O
(	O
see	O
top500	B-Operating_System
list	O
)	O
.	O
</s>
<s>
Although	O
most	O
computer	B-Architecture
clusters	I-Architecture
are	O
permanent	O
fixtures	O
,	O
attempts	O
at	O
flash	B-Device
mob	I-Device
computing	I-Device
have	O
been	O
made	O
to	O
build	O
short-lived	O
clusters	O
for	O
specific	O
computations	O
.	O
</s>
<s>
However	O
,	O
larger-scale	O
volunteer	B-Operating_System
computing	I-Operating_System
systems	O
such	O
as	O
BOINC-based	O
systems	O
have	O
had	O
more	O
followers	O
.	O
</s>
