Bruce Anderson wrote:
> 
> look in the connection log on the server and see what encoding
> the PC uses.
> 
> A couple of observations:
> 1) VNC server is based on old X11 code base.
> 2) server does all rendering in user space mem
>     on the server, where as in the case of X the
>     server runs on the client and has access to
>     video cards hardware accel.
> 3) See the mailing list archive for:
>    Subject: "VNCViewer slow on Win32 vs Linux"


Hi, Bruce,

I tried all encodings: tight, hextile raw,
RRE; all with copyrect, and switching the
8-bit-per-pixel on/off.  Compared to,
tightVNC, realVNC's automated parameter
selection won't help much since the
channel doesn't vary.

I think there are 2 levels of slowdown
happening here.  On the sun boxes, if both
the VNC server and viewer are 8-bit
pseudocolor, there is no detectable
delay.  If there is translation, there
is slight delay.  For example, I described
launching a 24-bit true color server in
which I launch a viewer to connect to the
8-bit pseudocolor server on which the
CAD tool is running.  This results in
tight encoding with compresslevel of 1
(though the compress level doesn't seem
to matter much).

Note that there are
actually two VNC links here, since I have
to connect to the 24-bit true color server
from an actual Xsession on the same machine.
The loopback connection's encoding doesn't
matter much either (raw, hextile, tight of
various levels).

So the end result is that translation
and dual VNC links causes a small bit of
slowdown, and the encoding doesn't matter
much.  So long as a PC is not involved.

The use of a PC for the viewer causes
significantly more slowdown, regardless of
encoding and whether 8-bit-per-pixel is
activated.  I noticed that the thread
you alluded to mentioned the encoding as
the possible cause.  However, these are
the same encodings that did not cause a
problem with the sun-to-sun connections.
I confirmed this by examining the server
log file in all cases.

So the cause must be either a bottleneck
between PC and sun networks; or, as implied
by you above (and in the thread), that the
PC-based viewer is not written to exploit
the PC's graphic rendering capabilities very
well.

For the case of narrow bandwidth, Exceed
might not be has hard hit by the narrow
bandwidth if it does make good use of PC
rendering abilities, and if the CAD app makes
use of high level X windows commands.  I'm
not familiar enough with X to know the extent
to which this can be done, if it is in fact
happening.  But I do observe that Exceed is
not as badly slowed down as VNC for the CAD
app.  The thread didn't mention which app,
but mine's is Cadence.

In addition to the possible causes above, the
fast flicking cursor I mentioned previously
hints at the possibility that there is
excessive polling or communication of some
sort between the app and the X server which
might be slowing down VNC.  Just
a (possibly naive) guess from a nongraphics
guy.  But it does corroborate with other
experiences.  It is only the IC layout
package that is hit so hard by slowness.
Other CAD packages are slowed, but still
usable.  For example, schematic capture is
usable, as is Synopsys synthesis.

Someone mentioned that their CAD package
inefficiently changes the colormap once
a second to make the cursor blink.  It may
or may not be related, but I think the slowdown
I'm seeing must be due to something more
severe than an event that happens once a second

I should add that VNC is
preferable for everything else but CAD.
This is because Exceed has this very annoying
delay in starting up new windows.  I can only
guess why.  Maybe it's doing a whole lot of X
book keeping as retribution for using high-
level commands.  Maybe  it's using some kind
of buffering scheme to paint a new frame.  As
I mentioned previously, it seems to switch
between two drastically different images very
fast.

Also, using Exceed, you run the risk of killing
all your work if the viewer dies.

Fred
-- 
Fred Ma, [EMAIL PROTECTED]
Carleton University, Dept. of Electronics
1125 Colonel By Drive, Ottawa, Ontario
Canada, K1S 5B6
_______________________________________________
VNC-List mailing list
[EMAIL PROTECTED]
To remove yourself from the list visit:
http://www.realvnc.com/mailman/listinfo/vnc-list

Reply via email to