<s>
Data-intensive	B-Operating_System
computing	I-Operating_System
is	O
a	O
class	O
of	O
parallel	B-Operating_System
computing	I-Operating_System
applications	O
which	O
use	O
a	O
data	B-Operating_System
parallel	I-Operating_System
approach	O
to	O
process	O
large	O
volumes	O
of	O
data	O
typically	O
terabytes	O
or	O
petabytes	O
in	O
size	O
and	O
typically	O
referred	O
to	O
as	O
big	B-Application
data	I-Application
.	O
</s>
<s>
Computing	O
applications	O
which	O
devote	O
most	O
of	O
their	O
execution	O
time	O
to	O
computational	O
requirements	O
are	O
deemed	O
compute-intensive	O
,	O
whereas	O
computing	O
applications	O
which	O
require	O
large	O
volumes	O
of	O
data	O
and	O
devote	O
most	O
of	O
their	O
processing	O
time	O
to	O
I/O	O
and	O
manipulation	O
of	O
data	O
are	O
deemed	O
data-intensive	B-Operating_System
.	O
</s>
<s>
In	O
addition	O
,	O
business	O
and	O
government	O
organizations	O
create	O
large	O
amounts	O
of	O
both	O
structured	O
and	O
unstructured	B-Application
information	I-Application
which	O
needs	O
to	O
be	O
processed	O
,	O
analyzed	O
,	O
and	O
linked	O
.	O
</s>
<s>
Data-intensive	B-Operating_System
computing	I-Operating_System
is	O
intended	O
to	O
address	O
this	O
need	O
.	O
</s>
<s>
Parallel	B-Operating_System
processing	I-Operating_System
approaches	O
can	O
be	O
generally	O
classified	O
as	O
either	O
compute-intensive	O
,	O
or	O
data-intensive	B-Operating_System
.	O
</s>
<s>
Parallel	B-Operating_System
processing	I-Operating_System
of	O
compute-intensive	O
applications	O
typically	O
involves	O
parallelizing	O
individual	O
algorithms	O
within	O
an	O
application	O
process	O
,	O
and	O
decomposing	O
the	O
overall	O
application	O
process	O
into	O
separate	O
tasks	O
,	O
which	O
can	O
then	O
be	O
executed	O
in	O
parallel	O
on	O
an	O
appropriate	O
computing	O
platform	O
to	O
achieve	O
overall	O
higher	O
performance	O
than	O
serial	O
processing	O
.	O
</s>
<s>
This	O
is	O
often	O
referred	O
to	O
as	O
task	B-Operating_System
parallelism	I-Operating_System
.	O
</s>
<s>
Data-intensive	B-Operating_System
is	O
used	O
to	O
describe	O
applications	O
that	O
are	O
I/O	O
bound	O
or	O
with	O
a	O
need	O
to	O
process	O
large	O
volumes	O
of	O
data	O
.	O
</s>
<s>
Parallel	B-Operating_System
processing	I-Operating_System
of	O
data-intensive	B-Operating_System
applications	I-Operating_System
typically	O
involves	O
partitioning	O
or	O
subdividing	O
the	O
data	O
into	O
multiple	O
segments	O
which	O
can	O
be	O
processed	O
independently	O
using	O
the	O
same	O
executable	O
application	O
program	O
in	O
parallel	O
on	O
an	O
appropriate	O
computing	O
platform	O
,	O
then	O
reassembling	O
the	O
results	O
to	O
produce	O
the	O
completed	O
output	O
data	O
.	O
</s>
<s>
The	O
greater	O
the	O
aggregate	O
distribution	O
of	O
the	O
data	O
,	O
the	O
more	O
benefit	O
there	O
is	O
in	O
parallel	B-Operating_System
processing	I-Operating_System
of	O
the	O
data	O
.	O
</s>
<s>
Data-intensive	B-Operating_System
processing	O
requirements	O
normally	O
scale	O
linearly	O
according	O
to	O
the	O
size	O
of	O
the	O
data	O
and	O
are	O
very	O
amenable	O
to	O
straightforward	O
parallelization	B-Operating_System
.	O
</s>
<s>
The	O
fundamental	O
challenges	O
for	O
data-intensive	B-Operating_System
computing	I-Operating_System
are	O
managing	O
and	O
processing	O
exponentially	O
growing	O
data	O
volumes	O
,	O
significantly	O
reducing	O
associated	O
data	O
analysis	O
cycles	O
to	O
support	O
practical	O
,	O
timely	O
applications	O
,	O
and	O
developing	O
new	O
algorithms	O
which	O
can	O
scale	O
to	O
search	O
and	O
process	O
massive	O
amounts	O
of	O
data	O
.	O
</s>
<s>
Computer	O
system	O
architectures	O
which	O
can	O
support	O
data	B-Operating_System
parallel	I-Operating_System
applications	O
were	O
promoted	O
in	O
the	O
early	O
2000s	O
for	O
large-scale	O
data	O
processing	O
requirements	O
of	O
data-intensive	B-Operating_System
computing	I-Operating_System
.	O
</s>
<s>
Data-parallelism	B-Operating_System
applied	O
computation	O
independently	O
to	O
each	O
data	O
item	O
of	O
a	O
set	O
of	O
data	O
,	O
which	O
allows	O
the	O
degree	O
of	O
parallelism	B-Operating_System
to	O
be	O
scaled	O
with	O
the	O
volume	O
of	O
data	O
.	O
</s>
<s>
The	O
key	O
issues	O
with	O
developing	O
applications	O
using	O
data-parallelism	B-Operating_System
are	O
the	O
choice	O
of	O
the	O
algorithm	O
,	O
the	O
strategy	O
for	O
data	O
decomposition	O
,	O
load	B-Application
balancing	I-Application
on	O
processing	O
nodes	O
,	O
message	B-Architecture
passing	I-Architecture
communications	O
between	O
nodes	O
,	O
and	O
the	O
overall	O
accuracy	O
of	O
the	O
results	O
.	O
</s>
<s>
The	O
development	O
of	O
a	O
data	B-Operating_System
parallel	I-Operating_System
application	O
can	O
involve	O
substantial	O
programming	O
complexity	O
to	O
define	O
the	O
problem	O
in	O
the	O
context	O
of	O
available	O
programming	O
tools	O
,	O
and	O
to	O
address	O
limitations	O
of	O
the	O
target	O
architecture	O
.	O
</s>
<s>
Information	B-General_Concept
extraction	I-General_Concept
from	O
and	O
indexing	O
of	O
Web	O
documents	O
is	O
typical	O
of	O
data-intensive	B-Operating_System
computing	I-Operating_System
which	O
can	O
derive	O
significant	O
performance	O
benefits	O
from	O
data	B-Operating_System
parallel	I-Operating_System
implementations	O
since	O
Web	O
and	O
other	O
types	O
of	O
document	O
collections	O
can	O
typically	O
then	O
be	O
processed	O
in	O
parallel	O
.	O
</s>
<s>
Design	O
of	O
data-intensive	B-Operating_System
computing	I-Operating_System
platforms	O
to	O
provide	O
high	O
levels	O
of	O
reliability	O
,	O
efficiency	O
,	O
availability	O
,	O
and	O
scalability	O
.	O
</s>
<s>
Pacific	O
Northwest	O
National	O
Labs	O
defined	O
data-intensive	B-Operating_System
computing	I-Operating_System
as	O
“	O
capturing	O
,	O
managing	O
,	O
analyzing	O
,	O
and	O
understanding	O
data	O
at	O
volumes	O
and	O
rates	O
that	O
push	O
the	O
frontiers	O
of	O
current	O
technologies	O
”	O
.	O
</s>
<s>
Data-intensive	B-Operating_System
computing	I-Operating_System
platforms	O
typically	O
use	O
a	O
parallel	B-Operating_System
computing	I-Operating_System
approach	O
combining	O
multiple	O
processors	O
and	O
disks	O
in	O
large	O
commodity	O
computing	B-Architecture
clusters	I-Architecture
connected	O
using	O
high-speed	O
communications	O
switches	O
and	O
networks	O
which	O
allows	O
the	O
data	O
to	O
be	O
partitioned	O
among	O
the	O
available	O
computing	O
resources	O
and	O
processed	O
independently	O
to	O
achieve	O
performance	O
and	O
scalability	O
based	O
on	O
the	O
amount	O
of	O
data	O
.	O
</s>
<s>
A	O
cluster	O
can	O
be	O
defined	O
as	O
a	O
type	O
of	O
parallel	O
and	O
distributed	B-Architecture
system	I-Architecture
,	O
which	O
consists	O
of	O
a	O
collection	O
of	O
inter-connected	O
stand-alone	O
computers	O
working	O
together	O
as	O
a	O
single	O
integrated	O
computing	O
resource	O
.	O
</s>
<s>
This	O
approach	O
to	O
parallel	B-Operating_System
processing	I-Operating_System
is	O
often	O
referred	O
to	O
as	O
a	O
“	O
shared	O
nothing	O
”	O
approach	O
since	O
each	O
node	O
consisting	O
of	O
processor	O
,	O
local	O
memory	O
,	O
and	O
disk	O
resources	O
shares	O
nothing	O
with	O
other	O
nodes	O
in	O
the	O
cluster	O
.	O
</s>
<s>
In	O
parallel	B-Operating_System
computing	I-Operating_System
this	O
approach	O
is	O
considered	O
suitable	O
for	O
data-intensive	B-Operating_System
computing	I-Operating_System
and	O
problems	O
which	O
are	O
“	O
embarrassingly	O
parallel	O
”	O
,	O
i.e.	O
</s>
<s>
These	O
types	O
of	O
data	O
processing	O
problems	O
are	O
inherently	O
adaptable	O
to	O
various	O
forms	O
of	O
distributed	B-Architecture
computing	I-Architecture
including	O
clusters	O
,	O
data	O
grids	O
,	O
and	O
cloud	B-Architecture
computing	I-Architecture
.	O
</s>
<s>
Several	O
common	O
characteristics	O
of	O
data-intensive	B-Operating_System
computing	I-Operating_System
systems	O
distinguish	O
them	O
from	O
other	O
forms	O
of	O
computing	O
:	O
</s>
<s>
To	O
achieve	O
high	O
performance	O
in	O
data-intensive	B-Operating_System
computing	I-Operating_System
,	O
it	O
is	O
important	O
to	O
minimize	O
the	O
movement	O
of	O
data	O
.	O
</s>
<s>
Newer	O
technologies	O
such	O
as	O
InfiniBand	B-Architecture
allow	O
data	O
to	O
be	O
stored	O
in	O
a	O
separate	O
repository	O
and	O
provide	O
performance	O
comparable	O
to	O
collocated	O
data	O
.	O
</s>
<s>
Data-intensive	B-Operating_System
computing	I-Operating_System
systems	O
utilize	O
a	O
machine-independent	O
approach	O
in	O
which	O
applications	O
are	O
expressed	O
in	O
terms	O
of	O
high-level	O
operations	O
on	O
data	O
,	O
and	O
the	O
runtime	O
system	O
transparently	O
controls	O
the	O
scheduling	O
,	O
execution	O
,	O
load	B-Application
balancing	I-Application
,	O
communications	O
,	O
and	O
movement	O
of	O
programs	O
and	O
data	O
across	O
the	O
distributed	B-Architecture
computing	I-Architecture
cluster	O
.	O
</s>
<s>
Data-intensive	B-Operating_System
computing	I-Operating_System
systems	O
are	O
designed	O
to	O
be	O
fault	O
resilient	O
.	O
</s>
<s>
The	O
inherent	O
scalability	O
of	O
the	O
underlying	O
hardware	O
and	O
software	B-Architecture
architecture	I-Architecture
.	O
</s>
<s>
Data-intensive	B-Operating_System
computing	I-Operating_System
systems	O
can	O
typically	O
be	O
scaled	O
in	O
a	O
linear	O
fashion	O
to	O
accommodate	O
virtually	O
any	O
amount	O
of	O
data	O
,	O
or	O
to	O
meet	O
time-critical	O
performance	O
requirements	O
by	O
simply	O
adding	O
additional	O
processing	O
nodes	O
.	O
</s>
<s>
The	O
number	O
of	O
nodes	O
and	O
processing	O
tasks	O
assigned	O
for	O
a	O
specific	O
application	O
can	O
be	O
variable	O
or	O
fixed	O
depending	O
on	O
the	O
hardware	O
,	O
software	O
,	O
communications	O
,	O
and	O
distributed	B-Protocol
file	I-Protocol
system	I-Protocol
architecture	O
.	O
</s>
<s>
A	O
variety	O
of	O
system	O
architectures	O
have	O
been	O
implemented	O
for	O
data-intensive	B-Operating_System
computing	I-Operating_System
and	O
large-scale	O
data	O
analysis	O
applications	O
including	O
parallel	O
and	O
distributed	O
relational	B-Application
database	I-Application
management	I-Application
systems	I-Application
which	O
have	O
been	O
available	O
to	O
run	O
on	O
shared	O
nothing	O
clusters	O
of	O
processing	O
nodes	O
for	O
more	O
than	O
two	O
decades	O
.	O
</s>
<s>
Several	O
solutions	O
have	O
emerged	O
including	O
the	O
MapReduce	B-Operating_System
architecture	O
pioneered	O
by	O
Google	B-Application
and	O
now	O
available	O
in	O
an	O
open-source	O
implementation	O
called	O
Hadoop	B-Application
used	O
by	O
Yahoo	B-Application
,	O
Facebook	B-Application
,	O
and	O
others	O
.	O
</s>
<s>
LexisNexis	B-Library
Risk	I-Library
Solutions	I-Library
also	O
developed	O
and	O
implemented	O
a	O
scalable	O
platform	O
for	O
data-intensive	B-Operating_System
computing	I-Operating_System
which	O
is	O
used	O
by	O
LexisNexis	B-Library
.	O
</s>
<s>
The	O
MapReduce	B-Operating_System
architecture	O
and	O
programming	O
model	O
pioneered	O
by	O
Google	B-Application
is	O
an	O
example	O
of	O
a	O
modern	O
systems	O
architecture	O
designed	O
for	O
data-intensive	B-Operating_System
computing	I-Operating_System
.	O
</s>
<s>
The	O
MapReduce	B-Operating_System
architecture	O
allows	O
programmers	O
to	O
use	O
a	O
functional	O
programming	O
style	O
to	O
create	O
a	O
map	O
function	O
that	O
processes	O
a	O
key	B-Application
–	I-Application
value	I-Application
pair	I-Application
associated	O
with	O
the	O
input	O
data	O
to	O
generate	O
a	O
set	O
of	O
intermediate	O
key	B-Application
–	I-Application
value	I-Application
pairs	I-Application
,	O
and	O
a	O
reduce	O
function	O
that	O
merges	O
all	O
intermediate	O
values	O
associated	O
with	O
the	O
same	O
intermediate	O
key	O
.	O
</s>
<s>
Since	O
the	O
system	O
automatically	O
takes	O
care	O
of	O
details	O
like	O
partitioning	O
the	O
input	O
data	O
,	O
scheduling	O
and	O
executing	O
tasks	O
across	O
a	O
processing	O
cluster	O
,	O
and	O
managing	O
the	O
communications	O
between	O
nodes	O
,	O
programmers	O
with	O
no	O
experience	O
in	O
parallel	B-Operating_System
programming	I-Operating_System
can	O
easily	O
use	O
a	O
large	O
distributed	B-Architecture
processing	I-Architecture
environment	O
.	O
</s>
<s>
The	O
programming	O
model	O
for	O
MapReduce	B-Operating_System
architecture	O
is	O
a	O
simple	O
abstraction	O
where	O
the	O
computation	O
takes	O
a	O
set	O
of	O
input	O
key	B-Application
–	I-Application
value	I-Application
pairs	I-Application
associated	O
with	O
the	O
input	O
data	O
and	O
produces	O
a	O
set	O
of	O
output	O
key	B-Application
–	I-Application
value	I-Application
pairs	I-Application
.	O
</s>
<s>
These	O
Map	O
tasks	O
perform	O
user-specified	O
computations	O
on	O
each	O
input	O
key	B-Application
–	I-Application
value	I-Application
pair	I-Application
from	O
the	O
partition	O
of	O
input	O
data	O
assigned	O
to	O
the	O
task	O
,	O
and	O
generates	O
a	O
set	O
of	O
intermediate	O
results	O
for	O
each	O
key	O
.	O
</s>
<s>
For	O
more	O
complex	O
data	O
processing	O
procedures	O
,	O
multiple	O
MapReduce	B-Operating_System
calls	O
may	O
be	O
linked	O
together	O
in	O
sequence	O
.	O
</s>
<s>
Apache	B-Application
Hadoop	I-Application
is	O
an	O
open	O
source	O
software	O
project	O
sponsored	O
by	O
The	O
Apache	O
Software	O
Foundation	O
which	O
implements	O
the	O
MapReduce	B-Operating_System
architecture	O
.	O
</s>
<s>
Hadoop	B-Application
now	O
encompasses	O
multiple	O
subprojects	O
in	O
addition	O
to	O
the	O
base	O
core	O
,	O
MapReduce	B-Operating_System
,	O
and	O
HDFS	O
distributed	O
filesystem	O
.	O
</s>
<s>
These	O
additional	O
subprojects	O
provide	O
enhanced	O
application	O
processing	O
capabilities	O
to	O
the	O
base	O
Hadoop	B-Application
implementation	O
and	O
currently	O
include	O
Avro	O
,	O
Pig	O
,	O
HBase	B-Language
,	O
ZooKeeper	B-Language
,	O
Hive	B-General_Concept
,	O
and	O
Chukwa	O
.	O
</s>
<s>
The	O
Hadoop	B-Application
MapReduce	B-Operating_System
architecture	O
is	O
functionally	O
similar	O
to	O
the	B-Application
Google	I-Application
implementation	O
except	O
that	O
the	O
base	O
programming	O
language	O
for	O
Hadoop	B-Application
is	O
Java	B-Language
instead	O
of	O
C++	B-Language
.	O
</s>
<s>
Hadoop	B-Application
implements	O
a	O
distributed	O
data	O
processing	O
scheduling	O
and	O
execution	O
environment	O
and	O
framework	O
for	O
MapReduce	B-Operating_System
jobs	O
.	O
</s>
<s>
Hadoop	B-Application
includes	O
a	O
distributed	B-Protocol
file	I-Protocol
system	I-Protocol
called	O
HDFS	O
which	O
is	O
analogous	O
to	O
GFS	B-Application
in	O
the	B-Application
Google	I-Application
MapReduce	B-Operating_System
implementation	O
.	O
</s>
<s>
The	O
Hadoop	B-Application
execution	O
environment	O
supports	O
additional	O
distributed	O
data	O
processing	O
capabilities	O
which	O
are	O
designed	O
to	O
run	O
using	O
the	O
Hadoop	B-Application
MapReduce	B-Operating_System
architecture	O
.	O
</s>
<s>
These	O
include	O
HBase	B-Language
,	O
a	O
distributed	O
column-oriented	O
database	O
which	O
provides	O
random	O
access	O
read/write	O
capabilities	O
;	O
Hive	B-General_Concept
which	O
is	O
a	O
data	O
warehouse	O
system	O
built	O
on	O
top	O
of	O
Hadoop	B-Application
that	O
provides	O
SQL-like	O
query	O
capabilities	O
for	O
data	O
summarization	O
,	O
ad	O
hoc	O
queries	O
,	O
and	O
analysis	O
of	O
large	O
datasets	O
;	O
and	O
Pig	O
–	O
a	O
high-level	O
data-flow	O
programming	O
language	O
and	O
execution	O
framework	O
for	O
data-intensive	B-Operating_System
computing	I-Operating_System
.	O
</s>
<s>
Pig	O
was	O
developed	O
at	O
Yahoo	B-Application
!	O
</s>
<s>
to	O
provide	O
a	O
specific	O
language	O
notation	O
for	O
data	O
analysis	O
applications	O
and	O
to	O
improve	O
programmer	O
productivity	O
and	O
reduce	O
development	O
cycles	O
when	O
using	O
the	O
Hadoop	B-Application
MapReduce	B-Operating_System
environment	O
.	O
</s>
<s>
Pig	O
programs	O
are	O
automatically	O
translated	O
into	O
sequences	O
of	O
MapReduce	B-Operating_System
programs	O
if	O
needed	O
in	O
the	O
execution	O
environment	O
.	O
</s>
<s>
HPCC	B-Operating_System
(	O
High-Performance	B-Operating_System
Computing	I-Operating_System
Cluster	I-Operating_System
)	O
was	O
developed	O
and	O
implemented	O
by	O
LexisNexis	B-Library
Risk	I-Library
Solutions	I-Library
.	O
</s>
<s>
The	O
HPCC	B-Operating_System
approach	O
also	O
utilizes	O
commodity	O
clusters	O
of	O
hardware	O
running	O
the	O
Linux	B-Application
operating	I-Application
system	I-Application
.	O
</s>
<s>
Custom	O
system	O
software	O
and	O
middleware	O
components	O
were	O
developed	O
and	O
layered	O
on	O
the	O
base	O
Linux	B-Application
operating	I-Application
system	I-Application
to	O
provide	O
the	O
execution	O
environment	O
and	O
distributed	O
filesystem	O
support	O
required	O
for	O
data-intensive	B-Operating_System
computing	I-Operating_System
.	O
</s>
<s>
LexisNexis	B-Library
also	O
implemented	O
a	O
new	O
high-level	O
language	O
for	O
data-intensive	B-Operating_System
computing	I-Operating_System
.	O
</s>
<s>
The	O
ECL	B-Language
programming	I-Language
language	I-Language
is	O
a	O
high-level	O
,	O
declarative	O
,	O
data-centric	O
,	O
implicitly	B-Operating_System
parallel	I-Operating_System
language	O
that	O
allows	O
the	O
programmer	O
to	O
define	O
what	O
the	O
data	O
processing	O
result	O
should	O
be	O
and	O
the	O
dataflows	O
and	O
transformations	O
that	O
are	O
necessary	O
to	O
achieve	O
the	O
result	O
.	O
</s>
<s>
The	O
ECL	B-Language
language	O
includes	O
extensive	O
capabilities	O
for	O
data	O
definition	O
,	O
filtering	O
,	O
data	O
management	O
,	O
and	O
data	O
transformation	O
,	O
and	O
provides	O
an	O
extensive	O
set	O
of	O
built-in	O
functions	O
to	O
operate	O
on	O
records	O
in	O
datasets	O
which	O
can	O
include	O
user-defined	O
transformation	O
functions	O
.	O
</s>
<s>
ECL	B-Language
programs	O
are	O
compiled	O
into	O
optimized	O
C++	B-Language
source	O
code	O
,	O
which	O
is	O
subsequently	O
compiled	O
into	O
executable	O
code	O
and	O
distributed	O
to	O
the	O
nodes	O
of	O
a	O
processing	O
cluster	O
.	O
</s>
<s>
To	O
address	O
both	O
batch	O
and	O
online	O
aspects	O
data-intensive	B-Operating_System
computing	I-Operating_System
applications	O
,	O
HPCC	B-Operating_System
includes	O
two	O
distinct	O
cluster	O
environments	O
,	O
each	O
of	O
which	O
can	O
be	O
optimized	O
independently	O
for	O
its	O
parallel	O
data	O
processing	O
purpose	O
.	O
</s>
<s>
A	O
Thor	O
system	O
is	O
similar	O
in	O
its	O
hardware	O
configuration	O
,	O
function	O
,	O
execution	O
environment	O
,	O
filesystem	O
,	O
and	O
capabilities	O
to	O
the	O
Hadoop	B-Application
MapReduce	B-Operating_System
platform	O
,	O
but	O
provides	O
higher	O
performance	O
in	O
equivalent	O
configurations	O
.	O
</s>
<s>
A	O
Roxie	O
system	O
is	O
similar	O
in	O
its	O
function	O
and	O
capabilities	O
to	O
Hadoop	B-Application
with	O
HBase	B-Language
and	O
Hive	B-General_Concept
capabilities	O
added	O
,	O
but	O
provides	O
an	O
optimized	O
execution	O
environment	O
and	O
filesystem	O
for	O
high-performance	O
online	O
processing	O
.	O
</s>
<s>
Both	O
Thor	O
and	O
Roxie	O
systems	O
utilize	O
the	O
same	O
ECL	B-Language
programming	I-Language
language	I-Language
for	O
implementing	O
applications	O
,	O
increasing	O
programmer	O
productivity	O
.	O
</s>
