[lug] It slows down

[D. Stimits]

A lot of people have mentioned some of the basic things you have to do
before any conclusions, like monitoring with top, and being suspicious
of rogue process after being rooted. But if the basics appear to be ok,
you might be interested to know that recently the topic of swap, cache,
and memory aging (and restated in some odd way, responsiveness) have
been a big debate in the 2.4.x kernels. Extreme changes have been going
on in that series, trying to solve a lot of problems. Pay close
attention to swap useage when it slows (assuming you have ruled out the
other problems first). Try a newer 2.4.x kernel, especially the Alan Cox
ac series. Patches:
ftp://zeus.kernel.org/pub/linux/kernel/people/alan/2.4

Before you actually install a newer kernel, check the source document:
/usr/src/linux/Documentation/Changes
Specifically, make sure all require versions are up to date (there is a
short list of packages and how to check version there).

It is possible you are running into the cache/swap/page aging issues.

D. Stimits, stimits at idcomm.com

David wrote:
> 
> Today I managed to capture some data on an intermittently occurring
> problem.  The problem is something that I have always realised I could
> be imagining; but today I measured convincing evidence that sometimes
> something strange happens to my machine.  I am reporting it in case
> anyone else has seen something similar, or knows about it, etc.
> Unfortunately, I do not have nearly enough data to deduce any
> conclusion with confidence; and intentionally doing experiments has
> been perceived as too much of a time imposition, at least up until
> now.
> 
> As a starting point I shall describe the problem as a significant
> slowing down of my machine from some moment in time.  Once the
> computer has "switched" into this mode everything takes much longer to
> run: I get spurious pauses, delays, and slow performance.  The datum I
> captured today is that, after the computer started to run slowly, a
> computationally intensive programme of mine took 7446 seconds to run.
> After this run had completed, I did a soft boot, and re-ran the
> programme without changing any data, i.e., an identical run from my
> point of view - it took 2607 seconds.
> 
> There are many more questions than answers here.  All I can do is
> itemise what may be relevant points.
> 
> * Currently I am running KRUD/RH7.0; but I have seen this problem
>   without KRUD and with earlier versions of RedHat.
> 
> * Normally I shut-down and turn-off my computer at night and re-boot
>   in the morning.  This was the case today.
> 
> * Pentium II, 200mHz, 128mB, IDE discs
> aside: I bought an AMD 1.4gHz, 0.5gB, 7200rpm machine today, which I
> pick up tomorrow; whoopee-do, I am hoping that my life is about to
> take off!
> 
> * Today was typical, I ran several copies of Emacs, three copies of
>   xdvi, netscape, and several xterms.  I used a calculator briefly.
> 
> * It is possible that the slowdown always occurs when the machine has
>   several, say ten, of these processes running simultaneously.
> 
> * It is possible that the slowdown occurs during use of netscape.
> 
> * Once the machine has flipped into go-slow mode it will not come out
>   of it just by killing processes; I am 90% sure of this, and I shall
>   monitor this more carefully in the future.
> 
> * There is no rogue process that I have detected by using ps.
> 
> * Could this problem be caused by swapping?  Note that when I start
>   one of these jobs I go away and do something else since I need the
>   results of one run to decide what to do next.  So, as far as I know,
>   there is nothing competing for cycles and no need for swapping.
>   Even if I do not go away normally all I do is edit text or something
>   else fairly benign.  Even if I compile something else, the
>   consequent slow down of the two processes has a different feel to it
>   than "the problem"; and, of course, the slow down goes away when one
>   of the processes finishes.
> 
> * Is the problem in:
> 
> + the hardware?
> 
> + Netscape?  I am suspicious of this, but I just do not know if I am
>   always running netscape when the flip occurs.  I shall monitor this
>   in the future.
> 
> + Java applets emitted from netscape?
> 
> + Emacs?  That would be a real find!!
> 
> + My code.  Well, this is a good question.  But I get different times
>   with identical runs.  I am biased, of course, but I doubt that this
>   is it.
> 
> + Window manager, fvwm.  This and netscape are my favorite suspects.
> 
> + Linux?  It could be...
> 
> Pretty weird, vague as well, I know.  However, it is fairly clear that
> something is being started, somehow, that eats cycles voraciously; and
> the lack of a rogue process (is there a better way to check than using
> ps?) points to an existing process going haywire.
> 
> So, this description is just for the record.  But if you have seen
> anything like this, then I. for one, will be interested to hear about
> it.
> _______________________________________________
> Web Page:  http://lug.boulder.co.us
> Mailing List: http://lists.lug.boulder.co.us/mailman/listinfo/lug