<s>
VirtualGL	O
is	O
an	O
open-source	B-Application
software	I-Application
package	O
that	O
redirects	O
the	O
3D	O
rendering	O
commands	O
from	O
Unix	B-Application
and	O
Linux	B-Application
OpenGL	B-Application
applications	O
to	O
3D	B-Architecture
accelerator	I-Architecture
hardware	O
in	O
a	O
dedicated	O
server	O
and	O
sends	O
the	O
rendered	O
output	O
to	O
a	O
(	O
thin	B-Device
)	O
client	O
located	O
elsewhere	O
on	O
the	O
network	O
.	O
</s>
<s>
Clients	O
can	O
connect	O
to	O
the	O
server	O
either	O
using	O
a	O
remote	B-Protocol
X11	I-Protocol
connection	I-Protocol
or	O
using	O
an	O
X11	B-Operating_System
proxy	O
such	O
as	O
a	O
VNC	B-Protocol
server	I-Protocol
.	O
</s>
<s>
In	O
case	O
of	O
an	O
X11	B-Operating_System
connection	O
some	O
client-side	O
VirtualGL	O
software	O
is	O
also	O
needed	O
to	O
receive	O
the	O
rendered	O
graphics	O
output	O
separately	O
from	O
the	O
X11	B-Operating_System
stream	O
.	O
</s>
<s>
In	O
case	O
of	O
a	O
VNC	B-Protocol
connection	O
no	O
specific	O
client-side	O
software	O
is	O
needed	O
other	O
than	O
the	O
VNC	B-Protocol
client	O
itself	O
.	O
</s>
<s>
The	O
performance	O
of	O
OpenGL	B-Application
applications	O
can	O
be	O
greatly	O
improved	O
by	O
rendering	O
the	O
graphics	O
on	O
dedicated	O
hardware	O
accelerators	O
that	O
are	O
typically	O
present	O
in	O
a	O
GPU	B-Architecture
.	O
</s>
<s>
GPUs	B-Architecture
have	O
become	O
so	O
commonplace	O
that	O
applications	O
have	O
come	O
to	O
rely	O
on	O
them	O
for	O
acceptable	O
performance	O
.	O
</s>
<s>
But	O
VNC	B-Protocol
and	O
other	O
thin	B-Device
client	I-Device
environments	O
for	O
Unix	B-Application
and	O
Linux	B-Application
do	O
not	O
have	O
access	O
to	O
such	O
hardware	O
on	O
the	O
server	O
side	O
.	O
</s>
<s>
Therefore	O
they	O
either	O
do	O
not	O
support	O
OpenGL	B-Application
applications	O
at	O
all	O
or	O
resort	O
to	O
slower	O
methods	O
such	O
as	O
rendering	O
on	O
the	O
client	O
or	O
in	O
software	O
on	O
the	O
server	O
.	O
</s>
<s>
Indirect	O
rendering	O
uses	O
the	O
GLX	B-Device
extension	O
to	O
the	O
X	B-Operating_System
Window	I-Operating_System
System	I-Operating_System
(	O
"	O
X11	B-Operating_System
"	O
or	O
"	O
X	B-Operating_System
"	O
)	O
to	O
encapsulate	O
the	O
OpenGL	B-Application
commands	O
inside	O
of	O
the	O
X11	B-Protocol
protocol	I-Protocol
stream	I-Protocol
and	O
ship	O
them	O
from	O
an	O
application	O
to	O
an	O
X	B-Operating_System
display	O
.	O
</s>
<s>
Traditionally	O
,	O
the	O
application	O
runs	O
on	O
a	O
remotely	O
located	O
application	O
server	O
,	O
and	O
the	O
X	B-Operating_System
display	O
runs	O
on	O
the	O
user	O
's	O
desktop	O
.	O
</s>
<s>
In	O
this	O
scenario	O
,	O
all	O
of	O
the	O
OpenGL	B-Application
commands	O
are	O
executed	O
by	O
the	O
user	O
's	O
desktop	O
machine	O
,	O
so	O
that	O
machine	O
must	O
have	O
a	O
fast	O
3D	O
graphics	B-Architecture
accelerator	I-Architecture
.	O
</s>
<s>
Many	O
OpenGL	B-Application
applications	O
,	O
however	O
,	O
do	O
not	O
meet	O
these	O
criteria	O
.	O
</s>
<s>
To	O
further	O
complicate	O
matters	O
,	O
some	O
OpenGL	B-Application
extensions	O
do	O
not	O
work	O
in	O
an	O
indirect	O
rendering	O
environment	O
.	O
</s>
<s>
In	O
other	O
cases	O
,	O
the	O
user	O
's	O
X	B-Operating_System
display	O
may	O
not	O
provide	O
explicit	O
support	O
for	O
a	O
needed	O
OpenGL	B-Application
extension	O
,	O
or	O
the	O
extension	O
may	O
rely	O
on	O
a	O
specific	O
hardware	O
configuration	O
that	O
is	O
not	O
present	O
on	O
the	O
user	O
's	O
desktop	O
machine	O
.	O
</s>
<s>
Performing	O
OpenGL	B-Application
rendering	O
on	O
the	O
application	O
server	O
circumvents	O
the	O
issues	O
introduced	O
by	O
indirect	O
rendering	O
,	O
since	O
the	O
application	O
now	O
has	O
a	O
fast	O
and	O
direct	O
path	O
to	O
the	O
3D	O
rendering	O
hardware	O
.	O
</s>
<s>
The	O
problem	O
then	O
becomes	O
how	O
to	O
stream	O
1-2	O
megapixels	O
of	O
image	O
data	O
over	O
a	O
network	O
at	O
interactive	O
frame	O
rates	O
,	O
but	O
commodity	O
technologies	O
(	O
HDTV	B-Device
,	O
to	O
name	O
one	O
)	O
already	O
address	O
this	O
problem	O
.	O
</s>
<s>
VirtualGL	O
uses	O
"	O
GLX	B-Device
forking	O
"	O
to	O
perform	O
OpenGL	B-Application
rendering	O
on	O
the	O
application	O
server	O
.	O
</s>
<s>
Unix	B-Application
and	O
Linux	B-Application
OpenGL	B-Application
applications	O
normally	O
send	O
both	O
GLX	B-Device
commands	O
and	O
ordinary	O
X11	B-Operating_System
commands	O
to	O
the	O
same	O
X	B-Operating_System
display	O
.	O
</s>
<s>
The	O
GLX	B-Device
commands	O
are	O
used	O
to	O
bind	O
OpenGL	B-Application
rendering	O
contexts	O
to	O
a	O
particular	O
X	B-Operating_System
window	I-Operating_System
,	O
obtain	O
a	O
list	O
of	O
pixel	O
formats	O
that	O
the	O
X	B-Operating_System
display	O
supports	O
,	O
etc	O
.	O
</s>
<s>
VirtualGL	O
takes	O
advantage	O
of	O
a	O
feature	O
in	O
Unix	B-Application
and	O
Linux	B-Application
that	O
allows	O
one	O
to	O
"	O
preload	O
"	O
a	O
library	O
into	O
an	O
application	O
,	O
effectively	O
intercepting	O
(	O
AKA	O
"	O
interposing	O
"	O
)	O
certain	O
function	O
calls	O
that	O
the	O
application	O
would	O
normally	O
make	O
to	O
shared	O
libraries	O
with	O
which	O
it	O
is	O
linked	O
.	O
</s>
<s>
Once	O
VirtualGL	O
is	O
preloaded	O
into	O
a	O
Unix	B-Application
or	O
Linux	B-Application
OpenGL	B-Application
application	O
,	O
it	O
intercepts	O
the	O
GLX	B-Device
function	O
calls	O
from	O
the	O
application	O
and	O
rewrites	O
them	O
such	O
that	O
the	O
corresponding	O
GLX	B-Device
commands	O
are	O
sent	O
to	O
the	O
application	O
server	O
's	O
X	B-Operating_System
display	O
(	O
the	O
"	O
3D	O
X	B-Operating_System
Server	O
"	O
)	O
,	O
which	O
presumably	O
has	O
a	O
3D	O
hardware	O
accelerator	O
attached	O
.	O
</s>
<s>
Thus	O
,	O
VirtualGL	O
prevents	O
GLX	B-Device
commands	O
from	O
being	O
sent	O
over	O
the	O
network	O
to	O
the	O
user	O
's	O
X	B-Operating_System
display	O
or	O
to	O
a	O
virtual	O
X	B-Operating_System
display	O
(	O
"	O
X	B-Operating_System
proxy	O
"	O
)	O
,	O
such	O
as	O
VNC	B-Protocol
,	O
that	O
does	O
not	O
support	O
GLX	B-Device
.	O
</s>
<s>
Meanwhile	O
,	O
the	O
rest	O
of	O
the	O
function	O
calls	O
from	O
the	O
application	O
,	O
including	O
the	O
ordinary	O
X11	B-Operating_System
commands	O
used	O
to	O
draw	O
the	O
application	O
's	O
user	O
interface	O
,	O
are	O
allowed	O
to	O
pass	O
through	O
VirtualGL	O
without	O
modification	O
.	O
</s>
<s>
Internally	O
,	O
VirtualGL	O
's	O
interposer	O
engine	O
also	O
maintains	O
a	O
map	O
of	O
windows	O
to	O
Pbuffers	O
,	O
matches	O
visual	O
attributes	O
between	O
the	O
destination	O
X	B-Operating_System
display	O
(	O
the	O
"	O
2D	O
X	B-Operating_System
Server	O
"	O
)	O
and	O
the	O
3D	O
X	B-Operating_System
Server	O
,	O
and	O
performs	O
a	O
variety	O
of	O
other	O
hashing	O
functions	O
to	O
assure	O
that	O
the	O
GLX	B-Device
redirection	O
is	O
seamless	O
.	O
</s>
<s>
But	O
essentially	O
,	O
once	O
the	O
OpenGL	B-Application
context	O
is	O
established	O
on	O
the	O
application	O
server	O
's	O
X	B-Operating_System
display	O
,	O
VirtualGL	O
gets	O
out	O
of	O
the	O
way	O
and	O
allows	O
all	O
subsequent	O
OpenGL	B-Application
commands	O
to	O
pass	O
through	O
unimpeded	O
to	O
the	O
application	O
server	O
's	O
3D	O
hardware	O
.	O
</s>
<s>
Thus	O
,	O
the	O
application	O
can	O
automatically	O
use	O
whatever	O
OpenGL	B-Application
features	O
and	O
extensions	O
are	O
provided	O
by	O
the	O
application	O
server	O
's	O
hardware	O
and	O
drivers	O
.	O
</s>
<s>
Apart	O
from	O
marshaling	O
GLX	B-Device
commands	O
and	O
managing	O
Pbuffers	O
,	O
VirtualGL	O
also	O
reads	O
back	O
the	O
rendered	O
pixels	O
at	O
the	O
appropriate	O
time	O
(	O
usually	O
by	O
monitoring	O
glXSwapBuffers( )	O
or	O
glFinish( )	O
)	O
and	O
then	O
draws	O
those	O
pixels	O
into	O
application	O
's	O
X	B-Operating_System
window	I-Operating_System
using	O
standard	O
X	B-Operating_System
image	O
drawing	O
commands	O
.	O
</s>
<s>
Since	O
VirtualGL	O
is	O
redirecting	O
the	O
GLX	B-Device
commands	O
away	O
from	O
the	O
2D	O
X	B-Operating_System
Server	O
,	O
it	O
can	O
be	O
used	O
to	O
add	O
accelerated	O
3D	O
support	O
to	O
X	B-Operating_System
proxies	O
(	O
such	O
as	O
VNC	B-Protocol
)	O
as	O
well	O
as	O
to	O
prevent	O
indirect	O
OpenGL	B-Application
rendering	O
from	O
occurring	O
when	O
using	O
a	O
remote	O
X	B-Operating_System
display	O
.	O
</s>
<s>
Using	O
VirtualGL	O
in	O
concert	O
with	O
VNC	B-Protocol
or	O
another	O
X	B-Operating_System
proxy	O
allows	O
multiple	O
users	O
to	O
simultaneously	O
run	O
3D	O
applications	O
on	O
a	O
single	O
application	O
server	O
and	O
multiple	O
clients	O
to	O
share	O
each	O
session	O
.	O
</s>
<s>
However	O
,	O
VNC	B-Protocol
and	O
its	O
ilk	O
are	O
tuned	O
to	O
handle	O
2D	O
applications	O
with	O
large	O
areas	O
of	O
solid	O
color	O
,	O
few	O
colors	O
,	O
and	O
few	O
inter-frame	O
differences	O
.	O
</s>
<s>
The	O
workload	O
generated	O
by	O
drawing	O
rendered	O
images	O
from	O
an	O
OpenGL	B-Application
application	O
into	O
an	O
X	B-Operating_System
window	I-Operating_System
is	O
essentially	O
the	O
same	O
workload	O
as	O
a	O
video	O
player	O
,	O
and	O
off-the-shelf	O
thin	B-Device
client	I-Device
software	O
typically	O
lacks	O
sufficiently	O
fast	O
image	O
codecs	B-General_Concept
to	O
be	O
able	O
to	O
handle	O
this	O
workload	O
with	O
interactive	O
frame	O
rates	O
.	O
</s>
<s>
TurboVNC	B-Application
and	O
TigerVNC	B-Language
are	O
offshoots	O
of	O
TightVNC	B-Language
that	O
accelerate	O
the	O
Tight	O
and	O
JPEG	O
encoding	O
,	O
in	O
part	O
by	O
using	O
libjpeg-turbo	O
,	O
a	O
SIMD-accelerated	O
version	O
of	O
libjpeg	B-Language
.	O
</s>
<s>
Both	O
projects	O
provide	O
VNC	B-Protocol
servers	I-Protocol
as	O
well	O
as	O
client	O
applications	O
.	O
</s>
<s>
TurboVNC	B-Application
was	O
developed	O
by	O
the	O
same	O
team	O
as	O
VirtualGL	O
.	O
</s>
<s>
On	O
100	B-Protocol
Megabit	I-Protocol
Ethernet	I-Protocol
networks	O
it	O
can	O
display	O
more	O
than	O
50	O
Megapixels/second	O
with	O
perceptually	O
lossless	O
image	O
quality	O
.	O
</s>
<s>
TurboVNC	B-Application
includes	O
further	O
optimizations	O
that	O
allow	O
it	O
to	O
display	O
10	O
–	O
12	O
Megapixels/second	O
over	O
a	O
5	O
Megabit	O
broadband	O
link	O
,	O
with	O
noticeably	O
less	O
but	O
usable	O
image	O
quality	O
.	O
</s>
<s>
TurboVNC	B-Application
also	O
extends	O
TightVNC	B-Language
to	O
include	O
client-side	O
double	O
buffering	O
and	O
other	O
features	O
targeted	O
at	O
3D	O
applications	O
,	O
such	O
as	O
the	O
ability	O
to	O
send	O
a	O
lossless	O
copy	O
of	O
the	O
screen	O
image	O
during	O
periods	O
of	O
inactivity	O
.	O
</s>
<s>
TurboVNC	B-Application
and	O
VirtualGL	O
are	O
used	O
by	O
the	O
Texas	O
Advanced	O
Computing	O
Center	O
at	O
University	O
of	O
Texas	O
at	O
Austin	O
to	O
allow	O
users	O
of	O
TeraGrid	B-Operating_System
to	O
remotely	O
access	O
the	O
3D	O
rendering	O
capabilities	O
of	O
the	O
Stampede	O
Visualization	O
Cluster	O
.	O
</s>
<s>
TigerVNC	B-Language
is	O
a	O
more	O
recent	O
fork	O
of	O
TightVNC	B-Language
that	O
provides	O
the	O
same	O
performance	O
as	O
TurboVNC	B-Application
but	O
performs	O
better	O
in	O
other	O
aspects	O
.	O
</s>
<s>
When	O
using	O
the	O
VGL	O
Transport	O
,	O
VirtualGL	O
compresses	O
the	O
rendered	O
3D	O
images	O
in	O
process	O
using	O
the	O
same	O
optimized	O
JPEG	O
codec	B-General_Concept
that	O
TurboVNC	B-Application
uses	O
.	O
</s>
<s>
The	O
VirtualGL	O
Client	O
is	O
responsible	O
for	O
decompressing	O
the	O
images	O
and	O
drawing	O
the	O
pixels	O
into	O
the	O
appropriate	O
X	B-Operating_System
window	I-Operating_System
.	O
</s>
<s>
Meanwhile	O
,	O
the	O
non-OpenGL	O
elements	O
of	O
the	O
application	O
's	O
display	O
are	O
sent	O
over	O
the	O
network	O
using	O
the	O
standard	O
remote	O
X11	B-Operating_System
protocol	O
and	O
rendered	O
on	O
the	O
client	O
machine	O
.	O
</s>
<s>
This	O
approach	O
requires	O
that	O
an	O
X	B-Operating_System
display	O
be	O
present	O
on	O
the	O
client	O
machine	O
,	O
and	O
the	O
reliance	O
upon	O
the	O
remote	O
X11	B-Operating_System
protocol	O
for	O
performing	O
2D	O
rendering	O
means	O
that	O
many	O
applications	O
will	O
perform	O
poorly	O
when	O
using	O
the	O
VGL	O
Transport	O
on	O
high-latency	O
networks	O
.	O
</s>
<s>
The	O
VGL	O
Transport	O
also	O
reduces	O
the	O
server	O
CPU	B-Device
load	O
,	O
since	O
the	O
2D	O
rendering	O
is	O
occurring	O
on	O
the	O
client	O
,	O
and	O
the	O
VGL	O
Transport	O
allows	O
advanced	O
OpenGL	B-Application
features	O
,	O
such	O
as	O
quad-buffered	O
stereo	O
,	O
to	O
be	O
used	O
.	O
</s>
<s>
The	O
developers	O
of	O
VirtualGL	O
envision	O
the	O
primary	O
users	O
of	O
the	O
VGL	O
Transport	O
to	O
be	O
laptop	O
users	O
with	O
an	O
802.11g	O
wireless	O
or	O
a	O
fast	B-Protocol
Ethernet	I-Protocol
connection	O
to	O
the	O
application	O
server	O
.	O
</s>
<s>
VirtualGL	O
and	O
TurboVNC	B-Application
were	O
core	O
components	O
of	O
the	O
Sun	B-Device
Visualization	I-Device
System	I-Device
product	O
from	O
Sun	O
Microsystems	O
,	O
which	O
was	O
discontinued	O
in	O
April	O
2009	O
.	O
</s>
<s>
The	O
two	O
open	O
source	O
packages	O
were	O
combined	O
with	O
a	O
closed	B-Application
source	I-Application
plugin	O
that	O
allowed	O
VirtualGL	O
to	O
send	O
compressed	O
images	O
to	O
Sun	B-Protocol
Ray	I-Protocol
thin	B-Device
clients	I-Device
and	O
another	O
closed	B-Application
source	I-Application
package	O
that	O
integrated	O
VirtualGL	O
with	O
Sun	B-Operating_System
Grid	I-Operating_System
Engine	I-Operating_System
,	O
providing	O
resource	O
management	O
and	O
scheduling	O
for	O
remote	O
3D	O
jobs	O
.	O
</s>
<s>
v4.x.x	O
of	O
NoMachine	O
supports	O
VirtualGL	O
to	O
allow	O
users	O
to	O
run	O
3D	O
applications	O
in	O
NoMachine	O
desktop	O
sessions	O
.	O
</s>
<s>
v2.1	O
of	O
the	O
Scalable	O
Visualization	O
Array	O
software	O
from	O
HP	O
includes	O
components	O
that	O
integrate	O
with	O
VirtualGL	O
and	O
TurboVNC	B-Application
,	O
allowing	O
3D	O
jobs	O
to	O
be	O
scheduled	O
on	O
and	O
remotely	O
displayed	O
from	O
a	O
visualization	O
cluster	O
.	O
</s>
<s>
v3.0.0	O
of	O
ThinLinc	B-Protocol
is	O
designed	O
to	O
work	O
in	O
conjunction	O
with	O
VirtualGL	O
.	O
</s>
