beta 1 laptop install and fix

Craig Ringer craig at postnewspapers.com.au
Fri Sep 26 15:05:19 UTC 2003


>>Any information on what may be behind the performance improvements?
>>[snip]
>>I'm interested, as much as anything to know whether it actually affects 
>>me (long-time XFCE4 user, don't like GNOME/KDE).
> 
> Any gains in interative "feel" between Red Hat Linux 9 and Beta2 are
> very likely kernel changes. A compiler change that sped everything up
> 10% would be close to a miracle, and wouldn't have a noticeable 
> effect on most operations. 

[snip]

Thanks for the info. I was rather curious, and appreciate the response.

BTW, I've noticed quite apalling performance penalties on all my 
machines that appear to stem from freetype/xft2. Setting GDK_USE_XFT=0 
can in many cases massively increase the percieved responsiveness of 
apps. It appears to be text rendering - while text can be very slow, 
image handling in, say, the GIMP remains as snappy as ever. Switching 
back to a virtual desktop (well, view anyway) with a maximised mozilla 
is a good way to tell - you can watch it (quickly) redraw. Disabling 
xft2 support in mozilla makes it too fast to see.

I've observed the issue in RH8, RH9, and the new beta. It happens on 
both local and network X11, with varying video chipsets, resolutions, 
and X server configurations. I see issues with all sorts of machines - a 
Pentium 133 thin client (S3 Trio), a 300MHz PC-104 NSC Geode (NSC 
video), my home Athlon XP 1.5GHz (NVidia GeForce4-4200 with NVidia.com 
drivers), and the Dual Xeon 2.4GHz (ATi Rage XL) I use at work. When you 
can watch the different parts of the mozilla window redraw (albeit 
quickly) on a dual Xeon, it strikes me that something is a bit funny.

What really bought it home, though, was doing a 'find /' in a 
full-screen gnome-terminal with GDK_USE_XFT=1, and comparing it to the 
time taken in a full-screen xterm. I expect performance penalties from 
AA fonts, but it _doubled_ the time taken to list the entire tree. I'd 
done a 'find / >&/dev/null' before each to make sure the directory tree 
was cached, and as I have 750MB of RAM I'd say it fit ;-) . Setting 
GDK_USE_XFT=0 and using gnome-terminal resulted in an intermediate 
result, but somewhat closer to the AA gnome-terminal than the xterm. 
Sorry, I don't have the exact numbers on hand but can come up with them 
again quickly (ish) if needed.

I've ended up disabling the use of XFT2 and freetype where possible on 
my LTSP server at work, to keep the client machines (P133s with 32Mb of 
RAM) reasonably responsive. These machines are fairly snappy normally - 
with an 800x600 display - but become quite sluggish when using things 
like the xft2-enabled mozilla builds, or even displaying _menus_ in gtk2 
apps when xft2 is being used. OpenOffice also appears to suffer, as do 
qt3 applications.

Note that I've tried XFree86 4.3 and didn't notice much, if any, 
improvement in performance with text handling.

So - any idea why the performance hit with the use of XFT2/freetype is 
so large? Any ideas on how to address the issue - or plans for fixing it?

Craig Ringer






More information about the fedora-test-list mailing list