<s>
Programming	B-Algorithm
with	I-Algorithm
Big	I-Algorithm
Data	I-Algorithm
in	I-Algorithm
R	I-Algorithm
(	O
pbdR	B-Algorithm
)	O
is	O
a	O
series	O
of	O
R	B-Language
packages	O
and	O
an	O
environment	O
for	O
statistical	B-Algorithm
computing	I-Algorithm
with	O
big	B-Application
data	I-Application
by	O
using	O
high-performance	O
statistical	O
computation	O
.	O
</s>
<s>
The	O
pbdR	B-Algorithm
uses	O
the	O
same	O
programming	O
language	O
as	O
R	B-Language
with	O
S3/S4	B-Language
classes	O
and	O
methods	O
which	O
is	O
used	O
among	O
statisticians	O
and	O
data	B-Application
miners	I-Application
for	O
developing	O
statistical	B-Algorithm
software	I-Algorithm
.	O
</s>
<s>
The	O
significant	O
difference	O
between	O
pbdR	B-Algorithm
and	O
R	B-Language
code	I-Language
is	O
that	O
pbdR	B-Algorithm
mainly	O
focuses	O
on	O
distributed	B-Operating_System
memory	I-Operating_System
systems	O
,	O
where	O
data	O
are	O
distributed	O
across	O
several	O
processors	O
and	O
analyzed	O
in	O
a	O
batch	B-General_Concept
mode	I-General_Concept
,	O
while	O
communications	O
between	O
processors	O
are	O
based	O
on	O
MPI	B-Application
that	O
is	O
easily	O
used	O
in	O
large	O
high-performance	B-Architecture
computing	I-Architecture
(	O
HPC	O
)	O
systems	O
.	O
</s>
<s>
R	B-Language
system	O
mainly	O
focuses	O
on	O
single	O
multi-core	B-Architecture
machines	O
for	O
data	O
analysis	O
via	O
an	O
interactive	O
mode	O
such	O
as	O
GUI	B-Application
interface	I-Application
.	O
</s>
<s>
Two	O
main	O
implementations	O
in	O
R	B-Language
using	O
MPI	B-Application
are	O
Rmpi	O
and	O
pbdMPI	O
of	O
pbdR	B-Algorithm
.	O
</s>
<s>
The	O
pbdR	B-Algorithm
built	O
on	O
pbdMPI	O
uses	O
SPMD	B-Operating_System
parallelism	I-Operating_System
where	O
every	O
processor	O
is	O
considered	O
as	O
worker	O
and	O
owns	O
parts	O
of	O
data	O
.	O
</s>
<s>
The	O
SPMD	B-Operating_System
parallelism	I-Operating_System
introduced	O
in	O
mid	O
1980	O
is	O
particularly	O
efficient	O
in	O
homogeneous	O
computing	O
environments	O
for	O
large	O
data	O
,	O
for	O
example	O
,	O
performing	O
singular	O
value	O
decomposition	O
on	O
a	O
large	O
matrix	O
,	O
or	O
performing	O
clustering	O
analysis	O
on	O
high-dimensional	O
large	O
data	O
.	O
</s>
<s>
On	O
the	O
other	O
hand	O
,	O
there	O
is	O
no	O
restriction	O
to	O
use	O
manager/workers	B-Operating_System
parallelism	I-Operating_System
in	O
SPMD	B-Operating_System
parallelism	I-Operating_System
environment	O
.	O
</s>
<s>
The	O
Rmpi	O
uses	O
manager/workers	B-Operating_System
parallelism	I-Operating_System
where	O
one	O
main	O
processor	O
(	O
manager	O
)	O
serves	O
as	O
the	O
control	O
of	O
all	O
other	O
processors	O
(	O
workers	O
)	O
.	O
</s>
<s>
The	O
manager/workers	B-Operating_System
parallelism	I-Operating_System
introduced	O
around	O
early	O
2000	O
is	O
particularly	O
efficient	O
for	O
large	O
tasks	O
in	O
small	O
clusters	B-Architecture
,	O
for	O
example	O
,	O
bootstrap	B-Application
method	I-Application
and	O
Monte	B-Algorithm
Carlo	I-Algorithm
simulation	I-Algorithm
in	O
applied	O
statistics	O
since	O
i.i.d.	O
</s>
<s>
The	O
idea	O
of	O
SPMD	B-Operating_System
parallelism	I-Operating_System
is	O
to	O
let	O
every	O
processor	O
do	O
the	O
same	O
amount	O
of	O
work	O
,	O
but	O
on	O
different	O
parts	O
of	O
a	O
large	O
data	O
set	O
.	O
</s>
<s>
For	O
example	O
,	O
a	O
modern	O
GPU	B-Architecture
is	O
a	O
large	O
collection	O
of	O
slower	O
co-processors	O
that	O
can	O
simply	O
apply	O
the	O
same	O
computation	O
on	O
different	O
parts	O
of	O
relatively	O
smaller	O
data	O
,	O
but	O
the	O
SPMD	B-Operating_System
parallelism	I-Operating_System
ends	O
up	O
with	O
an	O
efficient	O
way	O
to	O
obtain	O
final	O
solutions	O
(	O
i.e.	O
</s>
<s>
Programming	O
with	O
pbdR	B-Algorithm
requires	O
usage	O
of	O
various	O
packages	O
developed	O
by	O
pbdR	B-Algorithm
core	O
team	O
.	O
</s>
<s>
Among	O
these	O
packages	O
,	O
pbdMPI	O
provides	O
wrapper	O
functions	O
to	O
MPI	B-Application
library	O
,	O
and	O
it	O
also	O
produces	O
a	O
shared	B-Library
library	I-Library
and	O
a	O
configuration	O
file	O
for	O
MPI	B-Application
environments	O
.	O
</s>
<s>
All	O
other	O
packages	O
rely	O
on	O
this	O
configuration	O
for	O
installation	O
and	O
library	O
loading	O
that	O
avoids	O
difficulty	O
of	O
library	B-Library
linking	I-Library
and	O
compiling	O
.	O
</s>
<s>
All	O
other	O
packages	O
can	O
directly	O
use	O
MPI	B-Application
functions	O
easily	O
.	O
</s>
<s>
pbdSLAP	O
---	O
bundles	O
scalable	O
dense	O
linear	O
algebra	O
libraries	O
in	O
double	O
precision	O
for	O
R	B-Language
,	O
based	O
on	O
ScaLAPACK	B-Application
version	O
2.0.2	O
which	O
includes	O
several	O
scalable	O
linear	O
algebra	O
packages	O
(	O
namely	O
BLACS	O
,	O
PBLAS	B-Operating_System
,	O
and	O
ScaLAPACK	B-Application
)	O
.	O
</s>
<s>
Among	O
those	O
packages	O
,	O
the	O
pbdDEMO	O
package	O
is	O
a	O
collection	O
of	O
20+	O
package	O
demos	O
which	O
offer	O
example	O
uses	O
of	O
the	O
various	O
pbdR	B-Algorithm
packages	O
,	O
and	O
contains	O
a	O
vignette	O
that	O
offers	O
detailed	O
explanations	O
for	O
the	O
demos	O
and	O
provides	O
some	O
mathematical	O
or	O
statistical	O
insight	O
.	O
</s>
<s>
to	O
execute	O
the	O
code	O
where	O
Rscript	B-Language
is	O
one	O
of	O
command	O
line	O
executable	O
program	O
.	O
</s>
<s>
The	O
following	O
example	O
modified	O
from	O
pbdMPI	O
illustrates	O
the	O
basic	O
syntax	O
of	O
the	O
language	O
of	O
pbdR	B-Algorithm
.	O
</s>
<s>
Since	O
pbdR	B-Algorithm
is	O
designed	O
in	O
SPMD	B-Operating_System
,	O
all	O
the	O
R	B-Language
scripts	O
are	O
stored	O
in	O
files	O
and	O
executed	O
from	O
the	O
command	O
line	O
via	O
mpiexec	O
,	O
mpirun	O
,	O
etc	O
.	O
</s>
<s>
to	O
execute	O
the	O
code	O
where	O
Rscript	B-Language
is	O
one	O
of	O
command	O
line	O
executable	O
program	O
.	O
</s>
<s>
The	O
following	O
example	O
modified	O
from	O
pbdDEMO	O
illustrates	O
the	O
basic	O
ddmatrix	O
computation	O
of	O
pbdR	B-Algorithm
which	O
performs	O
singular	O
value	O
decomposition	O
on	O
a	O
given	O
matrix	O
.	O
</s>
<s>
to	O
execute	O
the	O
code	O
where	O
Rscript	B-Language
is	O
one	O
of	O
command	O
line	O
executable	O
program	O
.	O
</s>
