Subject: Re: UVM/other problems for desktop users in current?
To: NetBSD-current Discussion List <>
From: George Michaelson <>
List: current-users
Date: 12/18/2002 14:20:59
On Tue, 17 Dec 2002 22:49:35 -0500 (EST) (Greg A. Woods)

> [ On Wednesday, December 18, 2002 at 12:58:00 (+1000), George Michaelson
> wrote: ]
> > Subject: UVM/other problems for desktop users in current?
> >
> > Can anybody suggest why this happens?
> You're running some very large programs that causes a lot of very
> visible context switching and probably some paging too (especially when
> your "cvs upates" and compiles and such compete for I/O and metadata
> buffer space).

I'm running what a normal desktop user self-supporting can expect to have to
run. If the mix is 'bad' then its worth noting the interactive response on
other BSD variants (FreeBSD) isn't this bad, and on Linux is impercepable by
comparison, *for the same job mix*

> If your Xserver and very little else were running on a workstation, and
> all that other goo was running on some server somewhere, then things
> would likely be somewhat better.  At least then your mouse wouldn't
> freeze up.

Indeed. But, other OS appear to cope 'better' for this jobmix. This jobmix is
not that unusual.

> You can't expect to be able to context switch through a half dozen
> monster applications, all while running high-I/O and CPU loads for
> "background" tasks, and not get some "jumpiness" from the most visible
> one of them, especially if you're even the tiniest bit shy on RAM.
> (have you ever looked at what really happens when you scoot the mouse
> around so that it touches large parts of a whole bunch of busy windows,
> i.e. windows with lots of context sensitive areas?)

I realize this is subjective measures stuff, the thing is, its noticably
differently worse on this NetBSD. Which is a shame, because in so many other
ways its a better, more runnable release family.

> I can run multiple "cvs update", "make package", and bunches of other
> applications like cricket and netsaint and various SNMP tools, even with
> relatively little RAM (i.e. my build machine pages a fair bit) and on a
> system with quite slow disks (just a pair of IDE's), and I don't see any
> _serious_ degradation of performance.  I'm running kernels that do not
> have some of the newer sysctls for tuning the VM, though I have tuned
> the kernel directly somewhat, and I get amazingly good response time
> from my shell prompt even when the load average is up to 15 or 20 or so.

Right. So in not running 'new' kernels, is it possible that we're on the
'older is better' track here, the difference between -stable and -current and
I am just in that window between new things being released, and new things
being tuned?

> I've never really been happy with the performance of the Xserver and
> various X11 applications (not even xterm) if I'm running them all on the
> same machine where I'm trying to get some "real" computing and I/O
> activity accomplished (eg. compiling etc.), regardless of what OS it's
> on.

No. Well, I mean No, thats not my experience. There is an order of magnitude
difference in the end-user experience delay here between NetBSD on my box, and
FreeBSD on my box a few months ago, and Linux on the other identical box we
have in-hand. 

So, from the general "big file I/O makes it slow" I'm  honing in on "X is
really not good in the jobmix here". -Is this maybe something fixable in X, or
in the way I run X? I self-compiled from /usr/xsrc.

(btw, "go away" is as always a valid response. This is whining. I will stop,
 now its on record. If somebody finds some trivial mis-tunings later on in
 the default config, ya heard it here :-))