<s>
GPI-Space	B-Operating_System
is	O
a	O
parallel	O
programming	O
development	O
software	O
,	O
developed	O
by	O
the	O
Fraunhofer	O
Institute	O
for	O
Industrial	O
Mathematics	O
(	O
ITWM	O
)	O
.	O
</s>
<s>
The	O
main	O
concept	O
behind	O
the	O
software	O
is	O
separation	O
of	O
domain	O
and	O
HPC	B-Architecture
knowledge	O
and	O
leaving	O
each	O
part	O
to	O
the	O
respective	O
experts	O
while	O
the	O
GPI-Space	B-Operating_System
as	O
framework	O
integrates	O
both	O
parts	O
together	O
.	O
</s>
<s>
GPI-Space	B-Operating_System
is	O
making	O
use	O
of	O
GPI	B-Application
to	O
solve	O
big	B-Application
data	I-Application
problems	O
more	O
efficient	O
than	O
current	O
solutions	O
.	O
</s>
<s>
GPI-Space	B-Operating_System
was	O
first	O
introduced	O
in	O
a	O
domain-specific	O
version	O
for	O
geology	O
,	O
under	O
the	O
name	O
SDPA	O
(	O
Seismic	O
Development	O
and	O
Programming	O
Architecture	O
)	O
at	O
SEG	O
2010	O
in	O
Houston	O
.	O
</s>
<s>
GPI	B-Application
Space	O
comes	O
with	O
several	O
layers	O
,	O
that	O
make	O
up	O
the	O
core	O
of	O
the	O
parallel	O
programming	O
development	O
software	O
.	O
</s>
<s>
In	O
a	O
large	O
scale	O
HPC	B-Architecture
clusters	B-Architecture
,	O
these	O
can	O
be	O
heterogeneous	O
and	O
consist	O
of	O
traditional	O
compute	O
nodes	O
as	O
well	O
as	O
nodes	O
with	O
accelerator	O
cards	O
,	O
such	O
as	O
GPUs	B-Architecture
or	O
Intel	O
's	O
Xeon	B-General_Concept
Phi	I-General_Concept
.	O
</s>
<s>
The	O
workflow	B-Application
engine	I-Application
translates	O
instructions	O
from	O
an	O
existing	O
workflow	O
in	O
XML	B-Protocol
format	O
with	O
special	O
GPI-Space	B-Operating_System
tags	O
into	O
the	O
runtime	O
environments	O
internal	O
instructions	O
which	O
are	O
based	O
on	O
Petri	B-Operating_System
nets	I-Operating_System
.	O
</s>
<s>
HPC	B-Architecture
knowledge	O
and	O
experience	O
by	O
Fraunhofer	O
ITWM	O
's	O
Competence-Center	O
High-Performance	B-Architecture
Computing	I-Architecture
(	O
CC-HPC	O
)	O
is	O
an	O
essential	O
contributor	O
to	O
the	O
engine	O
's	O
capability	O
of	O
generating	O
highly	O
optimal	O
parallel	O
codes	O
.	O
</s>
<s>
All	O
computation	O
with	O
GPI-Space	B-Operating_System
can	O
be	O
done	O
using	O
a	O
fast	O
parallel	O
file	O
system	O
,	O
such	O
as	O
BeeGFS	B-Application
,	O
which	O
is	O
very	O
similar	O
to	O
other	O
Big	B-Application
Data	I-Application
solutions	O
available	O
.	O
</s>
<s>
But	O
beyond	O
this	O
,	O
GPI-Space	B-Operating_System
is	O
capable	O
of	O
doing	O
all	O
computation	O
in	O
memory	O
,	O
as	O
well	O
,	O
thus	O
omitting	O
the	O
higher	O
latencies	O
and	O
performance	O
bottlenecks	O
of	O
traditional	O
I/O	O
.	O
</s>
<s>
Using	O
Fraunhofer	O
GPI	B-Application
(	O
see	O
also	O
graphic	O
"	O
GPI	B-Application
Architecture	O
"	O
)	O
,	O
one	O
big	O
block	O
of	O
a	O
partitioned	O
global	O
address	O
space	O
is	O
dynamically	O
allocated	O
.	O
</s>
<s>
The	O
RDMA	B-General_Concept
capability	O
allows	O
for	O
fast	O
,	O
single	O
sided	O
communication	O
.	O
</s>
<s>
To	O
showcase	O
the	O
validity	O
of	O
the	O
GPI-Space	B-Operating_System
approach	O
,	O
Fraunhofer	O
first	O
introduced	O
it	O
as	O
part	O
of	O
the	O
Seismic	O
Development	O
and	O
Programming	O
Architecture	O
(	O
SDPA	O
)	O
during	O
SEG	O
2010	O
in	O
Houston	O
,	O
TX	O
to	O
the	O
community	O
.	O
</s>
<s>
Developers	O
at	O
the	O
CC-HPC	O
have	O
put	O
together	O
domain	O
specific	O
solutions	O
for	O
seismic	O
data	O
that	O
includes	O
:	O
</s>
