Negative caching

[Hal Burgiss]

On Fri, Jun 19, 2009 at 03:24:19AM -0700, Steve Lamb wrote:
> Hal Burgiss wrote:
> > I think I've figured out why recent Ubuntu has such abysmal performance. Very
> > cleverly over a period of time, it takes all the free memory and uses it to
> > store useless junk that you will never ever use. And then holds on to the
> > useless stuff as long as it possibly can. And then you are forced to
> > work out of swap.
> 
>      Uhm... no.  Simple as that.  Whatever your problem is has no relation to 
> caching or buffers.  At least if it is you certainly have quite a further 
> distance to go to prove it than what you've presented in your message.  The 
> cached pages, for all intents an purposes, can be considered free.  The 
> difference between a cached page and a free page is just a couple of bits in a 
> memory map.

Actually I was trying to blend a little humor with a bad situation that I
don't fully understand. I have been a Linux only user since Redhat4 (late
90's). The last version of windows I owned personally was 3.1. The current
Ubuntu (desktop) is the poorest performing Linux systems I have experienced in
that time. On Redhat, I have had desktop X sessions that lasted over a year.
That is unthinkable now. I am living in terms of weeks, before the system is
either so slow or slow flaky, a reboot is the best remedy. I have both home
and work systems on 9.04. And both are problematic. Both ran fine on 8.04. I
have a number of 8.04 servers that are also fine (but I never run X on
servers). 

[hal at admin1 clients]$ uptime
 07:58:31 up 353 days, 21:25,  2 users,  load average: 0.00, 0.00, 0.00

That's a 8.04 LAMP server that seems to run flawlessly.
 
>      Your speed settings have to the result of something else going on because 
> the caching that the kernel does, if it were the cause, would be evident to 
> more people with heavier usage patterns than you.  It is the most commonly 
> maligned function the kernel performs.  ;)

I am totally open to other causes/solutions. But what is a better explanation
of a previously happy system whose performance degrades consistently and
severely over time? And when left idle for long periods of time, goes
completely in the toilet, to the point of it taking seconds for windows to
show any response whatsover to mouse clicks, seconds for keystrokes to appear
in text terminals, and minutes for new windows to open? A reboot or a restart
of X will bring back "normal" responsiveness immediately, only to degrade
again in days/weeks. There is no sign of excessive CPU usage, and no single
program is taking excessive memory (at least as reported). It seems to have
all the earmarks of a really wacky swap problem. If not, its doing a good
imitation of one. 

I am also a victim of the intel chipset trainwreck, so that may be a
complicating factor (on my home system), but doesn't explain a slow,
consistent degration that eventually makes the system unusuable. The work
desktop does not degrade as badly and I use compiz effects on that one, so
that might be another complicating factor there. 

Thanks! 


-- 
Hal