<s>
Quasi-opportunistic	B-Operating_System
supercomputing	I-Operating_System
is	O
a	O
computational	O
paradigm	O
for	O
supercomputing	B-Architecture
on	O
a	O
large	O
number	O
of	O
geographically	O
disperse	B-Architecture
computers	I-Architecture
.	O
</s>
<s>
Quasi-opportunistic	B-Operating_System
supercomputing	I-Operating_System
aims	O
to	O
provide	O
a	O
higher	O
quality	O
of	O
service	O
than	O
opportunistic	B-Architecture
resource	I-Architecture
sharing	I-Architecture
.	O
</s>
<s>
The	O
quasi-opportunistic	O
approach	O
coordinates	O
computers	O
which	O
are	O
often	O
under	O
different	O
ownerships	O
to	O
achieve	O
reliable	O
and	O
fault-tolerant	B-General_Concept
high	O
performance	O
with	O
more	O
control	O
than	O
opportunistic	O
computer	B-Architecture
grids	I-Architecture
in	O
which	O
computational	O
resources	O
are	O
used	O
whenever	O
they	O
may	O
become	O
available	O
.	O
</s>
<s>
While	O
the	O
"	O
opportunistic	O
match-making	O
"	O
approach	O
to	O
task	O
scheduling	O
on	O
computer	B-Architecture
grids	I-Architecture
is	O
simpler	O
in	O
that	O
it	O
merely	O
matches	O
tasks	O
to	O
whatever	O
resources	O
may	O
be	O
available	O
at	O
a	O
given	O
time	O
,	O
demanding	O
supercomputer	B-Architecture
applications	O
such	O
as	O
weather	B-General_Concept
simulations	I-General_Concept
or	O
computational	O
fluid	O
dynamics	O
have	O
remained	O
out	O
of	O
reach	O
,	O
partly	O
due	O
to	O
the	O
barriers	O
in	O
reliable	O
sub-assignment	O
of	O
a	O
large	O
number	O
of	O
tasks	O
as	O
well	O
as	O
the	O
reliable	O
availability	O
of	O
resources	O
at	O
a	O
given	O
time	O
.	O
</s>
<s>
The	O
quasi-opportunistic	O
approach	O
enables	O
the	O
execution	O
of	O
demanding	O
applications	O
within	O
computer	B-Architecture
grids	I-Architecture
by	O
establishing	O
grid-wise	O
resource	O
allocation	O
agreements	O
;	O
and	O
fault	B-General_Concept
tolerant	I-General_Concept
message	B-Architecture
passing	I-Architecture
to	O
abstractly	O
shield	O
against	O
the	O
failures	O
of	O
the	O
underlying	O
resources	O
,	O
thus	O
maintaining	O
some	O
opportunism	O
,	O
while	O
allowing	O
a	O
higher	O
level	O
of	O
control	O
.	O
</s>
<s>
The	O
general	O
principle	O
of	O
grid	B-Architecture
computing	I-Architecture
is	O
to	O
use	O
distributed	B-Architecture
computing	I-Architecture
resources	O
from	O
diverse	O
administrative	O
domains	O
to	O
solve	O
a	O
single	O
task	O
,	O
by	O
using	O
resources	O
as	O
they	O
become	O
available	O
.	O
</s>
<s>
BOINC	B-Operating_System
,	O
developed	O
at	O
the	O
University	O
of	O
California	O
,	O
Berkeley	O
is	O
an	O
example	O
of	O
a	O
volunteer-based	B-Operating_System
,	O
opportunistic	O
grid	B-Architecture
computing	I-Architecture
system	O
.	O
</s>
<s>
The	O
applications	O
based	O
on	O
the	O
BOINC	B-Operating_System
grid	O
have	O
reached	O
multi-petaflop	O
levels	O
by	O
using	O
close	O
to	O
half	O
a	O
million	O
computers	O
connected	O
on	O
the	O
internet	O
,	O
whenever	O
volunteer	O
resources	O
become	O
available	O
.	O
</s>
<s>
Another	O
system	O
,	O
Folding	B-Application
@home	I-Application
,	O
which	O
is	O
not	O
based	O
on	O
BOINC	B-Operating_System
,	O
computes	O
protein	O
folding	O
,	O
has	O
reached	O
8.8	O
petaflops	O
by	O
using	O
clients	O
that	O
include	O
GPU	B-Architecture
and	O
PlayStation	B-Operating_System
3	I-Operating_System
systems	O
.	O
</s>
<s>
However	O
,	O
these	O
results	O
are	O
not	O
applicable	O
to	O
the	O
TOP500	B-Operating_System
ratings	O
because	O
they	O
do	O
not	O
run	O
the	O
general	O
purpose	O
Linpack	B-Application
benchmark	O
.	O
</s>
<s>
A	O
key	O
strategy	O
for	O
grid	B-Architecture
computing	I-Architecture
is	O
the	O
use	O
of	O
middleware	B-General_Concept
that	O
partitions	O
pieces	O
of	O
a	O
program	O
among	O
the	O
different	O
computers	O
on	O
the	O
network	O
.	O
</s>
<s>
Although	O
general	O
grid	B-Architecture
computing	I-Architecture
has	O
had	O
success	O
in	O
parallel	O
task	O
execution	O
,	O
demanding	O
supercomputer	B-Architecture
applications	O
such	O
as	O
weather	B-General_Concept
simulations	I-General_Concept
or	O
computational	O
fluid	O
dynamics	O
have	O
remained	O
out	O
of	O
reach	O
,	O
partly	O
due	O
to	O
the	O
barriers	O
in	O
reliable	O
sub-assignment	O
of	O
a	O
large	O
number	O
of	O
tasks	O
as	O
well	O
as	O
the	O
reliable	O
availability	O
of	O
resources	O
at	O
a	O
given	O
time	O
.	O
</s>
<s>
The	O
opportunistic	O
supports	O
GIMPS	O
,	O
one	O
of	O
the	O
earliest	O
grid	B-Architecture
computing	I-Architecture
projects	O
since	O
1997	O
,	O
researching	O
Mersenne	O
prime	O
numbers	O
.	O
</s>
<s>
,	O
GIMPS	O
's	O
distributed	O
research	O
currently	O
achieves	O
about	O
60	O
teraflops	O
as	O
an	O
volunteer-based	B-Operating_System
computing	O
project	O
.	O
</s>
<s>
The	O
use	O
of	O
computing	O
resources	O
on	O
"	O
volunteer	B-Operating_System
grids	I-Operating_System
"	O
such	O
as	O
GIMPS	O
is	O
usually	O
purely	O
opportunistic	O
:	O
geographically	O
disperse	O
distributively	O
owned	O
computers	O
are	O
contributing	O
whenever	O
they	O
become	O
available	O
,	O
with	O
no	O
preset	O
commitments	O
that	O
any	O
resources	O
will	O
be	O
available	O
at	O
any	O
given	O
time	O
.	O
</s>
<s>
Furthermore	O
,	O
users	O
will	O
find	O
it	O
exceedingly	O
costly	O
to	O
organize	O
a	O
very	O
large	O
number	O
of	O
opportunistic	O
computing	O
resources	O
in	O
a	O
manner	O
that	O
can	O
achieve	O
reasonable	O
high	B-Architecture
performance	I-Architecture
computing	I-Architecture
.	O
</s>
<s>
An	O
example	O
of	O
a	O
more	O
structured	O
grid	O
for	O
high	B-Architecture
performance	I-Architecture
computing	I-Architecture
is	O
DEISA	O
,	O
a	O
supercomputer	B-Architecture
project	O
organized	O
by	O
the	O
European	O
Community	O
which	O
uses	O
computers	O
in	O
seven	O
European	O
countries	O
.	O
</s>
<s>
DEISA	O
has	O
a	O
two	O
level	O
integration	O
scheme	O
:	O
the	O
"	O
inner	O
level	O
"	O
consists	O
of	O
a	O
number	O
of	O
strongly	O
connected	O
high	B-Architecture
performance	I-Architecture
computer	I-Architecture
clusters	O
that	O
share	O
similar	O
operating	O
systems	O
and	O
scheduling	O
mechanisms	O
and	O
provide	O
a	O
homogeneous	O
computing	O
environment	O
;	O
while	O
the	O
"	O
outer	O
level	O
"	O
consists	O
of	O
heterogeneous	O
systems	O
that	O
have	O
supercomputing	B-Architecture
capabilities	O
.	O
</s>
<s>
Thus	O
DEISA	O
can	O
provide	O
somewhat	O
controlled	O
,	O
yet	O
dispersed	O
high	B-Architecture
performance	I-Architecture
computing	I-Architecture
services	O
to	O
users	O
.	O
</s>
<s>
Quasi-opportunistic	O
distributed	O
execution	O
of	O
demanding	O
parallel	O
computing	O
software	O
in	O
grids	O
focuses	O
on	O
the	O
implementation	O
of	O
grid-wise	O
allocation	O
agreements	O
,	O
co-allocation	O
subsystems	O
,	O
communication	O
topology-aware	O
allocation	O
mechanisms	O
,	O
fault	B-General_Concept
tolerant	I-General_Concept
message	B-Architecture
passing	I-Architecture
libraries	O
and	O
data	O
pre-conditioning	O
.	O
</s>
<s>
In	O
this	O
approach	O
,	O
fault	B-General_Concept
tolerant	I-General_Concept
message	B-Architecture
passing	I-Architecture
is	O
essential	O
to	O
abstractly	O
shield	O
against	O
the	O
failures	O
of	O
the	O
underlying	O
resources	O
.	O
</s>
<s>
The	O
quasi-opportunistic	O
approach	O
goes	O
beyond	O
volunteer	B-Operating_System
computing	I-Operating_System
on	O
a	O
highly	O
distributed	B-Architecture
systems	I-Architecture
such	O
as	O
BOINC	B-Operating_System
,	O
or	O
general	O
grid	B-Architecture
computing	I-Architecture
on	O
a	O
system	O
such	O
as	O
Globus	O
by	O
allowing	O
the	O
middleware	B-General_Concept
to	O
provide	O
almost	O
seamless	O
access	O
to	O
many	O
computing	B-Architecture
clusters	I-Architecture
so	O
that	O
existing	O
programs	O
in	O
languages	O
such	O
as	O
Fortran	O
or	O
C	O
can	O
be	O
distributed	O
among	O
multiple	O
computing	O
resources	O
.	O
</s>
<s>
A	O
key	O
component	O
of	O
the	O
quasi-opportunistic	O
approach	O
,	O
as	O
in	O
the	O
Qoscos	B-Device
Grid	I-Device
,	O
is	O
an	O
economic-based	O
resource	O
allocation	O
model	O
in	O
which	O
resources	O
are	O
provided	O
based	O
on	O
agreements	O
among	O
specific	O
supercomputer	B-Architecture
administration	O
sites	O
.	O
</s>
<s>
However	O
,	O
"	O
tit-for-tat	O
"	O
paradigms	O
in	O
which	O
computations	O
are	O
paid	O
back	O
via	O
future	O
computations	O
is	O
not	O
suitable	O
for	O
supercomputing	B-Architecture
applications	O
,	O
and	O
is	O
avoided	O
.	O
</s>
<s>
The	O
other	O
key	O
component	O
of	O
the	O
quasi-opportunistic	O
approach	O
is	O
a	O
reliable	O
message	B-Architecture
passing	I-Architecture
system	O
to	O
provide	O
distributed	O
checkpoint	B-General_Concept
restart	I-General_Concept
mechanisms	O
when	O
computer	O
hardware	O
or	O
networks	O
inevitably	O
experience	O
failures	O
.	O
</s>
