At 23.26 05/04/2003, you wrote:

> Is this necessary? It takes about 30 seconds or more!

there's something else wrong, it only takes about 3 seconds on my system
after the first run.

Could I speed it up by deleting modules I know are completely useless (now and in the future)?


>
> > > Unfortunately, when the login manager should have appeared, everything
> > > locked up again. I could hear the HD reading something, but I wasn't able
> > > to kill xfree because I could not access the terminals. Hard reboot AGAIN.
> > > I booted into runlevel 3 and executed XFdrake, this time trying to use a
> > > different driver for the video card. I didn't find one, but I discovered I
> > > could select xfree 4.3 wirthout 3d support or even xfree 3.3.6. I don't
> > > know why but the problem was 3D support: the Banshee card has 3d support
> > > with xfree, but who knows...
> >
> >Glad I saw this before trying to upgrade either of my servers -- they
> >both have Voodoo3 AGP cards in them. Have you tried using 3.3.6? It was
> >always pretty solid with the Voodoo.
>
> Thank you for the tip! I tried 3.3.6 but, since 4.3 worked, I chose the
> latest. Anyway, has 3.3.6 3d support? if not, where's the advantage?


you can call >1 second response time after clicking a menu working,
personally I'd call it "format the disk and try again." 3.3.6 does have
3d support, I used to play Quake 3 on it all the time. Sniff. I'd play
Quake 3 on my laptop now if it didn't insist on mmap'ing /dev/dsp.

I decided to make anotherm attempt and ... it worked! Now I can run xfree 4.3 with 3d acceleration. I will try 3.3.6 too, I wonder whether it could speed up apps.


A question: I have characters wrapped up in text consoles and I found this hint:

For 2.4 kernel, you also must disable console acceleration, so arguments are (for example)
'video=tdfx:[EMAIL PROTECTED] ,noaccel). Without that the characters will all be wrapped
up, making console useless (but X should be fine).

Do you ever applied this hint? have you ever had such problems? I also found this hint to enable tdfx framebuffer:

mknod /dev/fb0 c 29 0

How to know it's correct? I don't remember where I found it.


> > > Another strange thing: not only kde and gnome are slower than mdk 9.0, but
> > > even the whole system: for example, window maker or a simple "./configure"
> > > executed in a separate shell outside xfree (runlevel 5). Could it depends
> > > on the kernel? I don't know, but I find this situation very strange and
> > > unacceptable (I mean: I will come back to mdk 9.0 and I'll try to upgrade
> > > manually as much as I can).
> > >
> >
> >X stuff being slow could be an xserver problem, but this sounds like
> >something different... I don't know what though.
> >/etc/sysconfig/harddisks? Was the system responsive before?
>
> The system has always been responsive ... as responsive as a cel400 with a
> very poor chipset can be. This time, with 9.1, it's MUCH worse.
> The problem should not be related to the harddisk: the ibm hd is now pio (I
> will change it to dma soon), but the disk works only for few seconds, while
> apps take many many seconds to start, after the hd has stopped working. I
> forgot to tell that I set the hd in quietest mode, but this affects only
> seek time (so it could probably slows down the depmod process), while, as
> already said, apps take much to start even after the hd has finished.
>


well something is way wrong, but I don't know what it is.

Anyway, I checked: even if dmesg tells hda is in pio mode, hdparm tells hda is in udma4 mode, so everything is ok.
I also tried to unmask irq, but nothing changed. I get 37 MB/s from linux cache and 17-19 MB/s when reading from disk (contiguous data...), I wonder whether changing the setting to udma from 66 to 33 could speed up things. After all, now I barely saturate a pio mode! (rotten chipset! and this time is not mdk 9.1's fault).


> > > A question: why did you choose a -pre kernel? was it really necessary?
> > >
> >I'll hazard a guess that it's ACPI -- those of us with new laptops and
> >desktops don't get working systems without good ACPI support, and 2.4.21
> >is definitely a lot better.
>
> What about preemptive patch? is it already inside?
>

Not yet, but I would expect that the 9.2 distro will have it (if it
isn't a 2.6 kernel)

Could kernel 2.5.66 help somewhere? or compiling 2.4.20 or 21-preX only with specific modules? I have a very "special" box, maybe this time recompiling will help.


Thanks
Olaf


<olaf@ kjws.com> for every kind of mail, except spam! :-)



Want to buy your Pack or Services from MandrakeSoft? 
Go to http://www.mandrakestore.com

Reply via email to