Re: Poor X performance

On 1/13/08, cwillu <cwillu@cwillu.com> wrote:
> > First can someone suggest a better quantitative test for X 2d speed?
> > If I can measure something, it makes testing so much easier. Next,
> > can anyone suggest some other bits to toggle?
>
> Forgot to mention: x11perf might be useful.

Now the new ltsp does use ssh, which slows things down a bit. I
expect higher latency and cpu. However those don't seem to be related
to the most pronounced issue - scrolling through the inbox in
thunderbird. ssh usually only takes a small portion of the cpu, and
very little while scrolling in thunderbird.

x11perf definitely shows the new server to be slower. In some cases
substantially. For instance look at the 100x100 rectangle times:

Sync time adjustment is 1.8216 msecs.
0.0011 msec (914000.0/sec): Dot
0.0021 msec (473000.0/sec): 1x1 rectangle
0.2692 msec ( 3710.0/sec): 100x100 rectangle

Sync time adjustment is 1.1762 msecs.
0.0006 msec (1580000.0/sec): Dot
0.0014 msec (719000.0/sec): 1x1 rectangle
0.0009 msec (1080000.0/sec): 100x100 rectangle

So at least now I have a solid test to tweak things. I think I'll get
rid of ssh just to be sure. After that I'm not sure whats next.
Perhaps I'll try gusty.

Thanks for the help.
Received on Mon Jan 14 13:27:21 2008