Re: bad performance

[Chad Cunningham <ccunning math ohio-state edu>]

> This really does not sound like a Gnome problem -- I would expect you to
> have the same problems with any other environment (i.e. a std window
> manager, KDE, whatever). It sounds like it could be one of three
> possible things:

But I don't... Thus my confusion. I've been running helix+gnome since the
first betas and I'm running current releases on several other machines
currently. None of the other machines experience this problem, so I can't
really blame gnome. But, performance of my machine is excellent and the
only thing that exhibits this are gnome apps. And it's the same regardless
of which window manager I use with gnome. So I can't really blame a
hardware problem or the window manager. And this is a fairly stock install
of RedHat 6.2 which is set up exactly like several other machines I have,
so I can't blame the OS or setup.

So I'm a bit stuck here having nothing to blame. the only thing I can
think is that I either have a hardware problem that is only triggered by
something gnome does, or gnome has a problem that is triggered by
something in my specific hardware setup. Either that, or sunspots...

> 1 - Misconfigured X server. Run x11perf -all on both your new machine
> and old machine, and compare the numbers. I'd guess that you'll get
> slower numbers on your Athlon, which is NOT correct, since it is a much
> faster machine with a much faster video card. Are you sure that you have
> acceleration for your matrox enabled? (Such as, is the mouse pointer is
> being done in hardware?) Look at X's output when it starts up, check for
> any errors from the mga driver. (Run startx from a console.) Make sure
> you ARE getting the mga driver, not just a non-accel svga driver.

x11perf returns much better numbers for my 850/G400 than the 250/Permedia2
as one would expect. I haven't tweaked X much, I'm just using the standard
SVGA server with a standard configuration, which seems to work fine for
windowmaker, kde, fvrm, afterstep, ....

> 2 - Misconfigured kernel. Are you using a non-stable kernel? In
> particular, lots of the 2.3.99-preX series (notably, 6 and above) had
> major memory management issues; the resulting trash could cause you to
> see some of the symptoms you describe. If you aren't already, try
> running 2.2.15 or .16 (I can't vouch for .16, I have not ran it myself,
> but I trust Alan ;-) Check the output of 'dmesg' every once in a while
> and see if anything is spewing lost interrupts, IDE resets, or whatever.

Currently running the stock 2.2.14 redhat kernel, there are no intereting
log entries or errors anywhere...

> 3 - Hardware issues. You could be falling prey to one of the banes of PC
> hardware, namely interrupt sharing and the like. Look at
> /proc/interrupts. The G400 really wants its own interrupt. You can do
> this by installing it in, say, the topmost slot on your motherboard, and
> starting all other cards from the bottom. It especially hates sharing
> interrupts with other high-interrupt devices, like network cards. This
> could also cause some of the problems you indicate. Also, do you have
> the correct RAM? Athlon's are VERY picky about the kind of SDRAM you
> use. (Even though it says it's the right kind, voltage, timing, etc., it
> STILL might not work. I had to go through two different manufacturers
> before I found one that worked. Check AMD's pages to find a list of
> compatible vendors.) Having non-fully-compatible SDRAM installed could
> cause some of the issues you see.

My interrupts are fine, I'm certain it's not a ram problem as I would be
experiencing problems with things other than gnome.

> Speaking of IDE, you should check the hdparm settings for your drives
> (if you're not running SCSI). While it should not cause the problems you
> describe, making sure DMA is turned on, as well as multiple-sector
> access and a few other goodies can greatly speed up your system and
> stability.

I'm running a 7200 rpm ultradma drive with udma and 32 bit transfers
enabled for a throughput of about 23.4 MB/s, so disk speed is not the
problem.


-- 

Chad Cunningham
ccunning@math.ohio-state.edu

Only 102 days until RSA Patent expiration!