On Wed, 11 May 2011 14:28:35 -0500
DRC <dcomman...@users.sourceforge.net> wrote:
> I really don't understand at all what you're proposing. How is the
> quality adjusted in response to BWmax?
if (BW > 95% BWmax)
quality -= 10
else if (BW < 90% BWmax)
quality += 2
(Decrease quality quickly so as to converge on a usable framerate, then
increase slowly.)
> It seems to converge way too slowly, and I don't like the fact that it will
> send a higher-quality
> frame every 30 seconds (that will cause a stutter.)
It does not converge very quickly (something like 30 seconds I would say), but
it goes faster than the manual approach "let's try to lower the quality without
making the image unreadable". Also, even when it hasn't converged the
application is useable.
As far as the higher quality image is concerned, "higher" is quite a relative
notion - the idea is to force a stronger load on the network, but the higher
quality doesn't need to be 95, it can very well be current_quality + 10 (that's
what I have on my prototype IIRC). For what it's worth you don't even need to
raise the quality if you can live with some underestimation of BWmax.
I do think that the BWmax estimation I came up with is the most simple that can
be done, and assuming it gives decent results, it's a good basis to work upon.
If you don't like the higher quality image we can make do without it -
theoretically you will want to increase the quality a little bit to make sure
you make use of as most of the available resources as possible, but once you've
reached your image quality when really matters is to be able to re-evaluate
BWmax when it goes *down*, not *up*, and you don't need to increase the quality
to be able to reevaluate BWmax downwards every 30s.
(I'm aiming for something that is simple and works well enough, for the sake of
maintainability. If you don't maintain it I might have to do it myself so I
understand your concern very well.)
--
Greetings,
A. Huillet
------------------------------------------------------------------------------
Achieve unprecedented app performance and reliability
What every C/C++ and Fortran developer should know.
Learn how Intel has extended the reach of its next-generation tools
to help boost performance applications - inlcuding clusters.
http://p.sf.net/sfu/intel-dev2devmay
_______________________________________________
VirtualGL-Devel mailing list
VirtualGL-Devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/virtualgl-devel
