2.2.11 unstable

[Robert Dinse <nanook@eskimo.com>]

> From: Jon Masters <mastersj@periscope-systems.freeserve.co.uk>
> Date: Tue, 24 Aug 1999 04:22:28 +0100
> Subject: 2.2.11 unstable?
> 
> Hi,
>     I don't want you guys to take this the wrong way, but since I
> upgraded to 2.2.11, my machine has had more attacks of "instability"
> than an old winblows 95 pentium which my mother uses to do word
> processing. I am used to the odd crash or two - it's only natural
> (personally, I *never* blame the kernel, it's always gotta be a power
> spike or alpha particle - Linux would never normally crash, it's just
> too damm good :). However, over the last couple of weeks since I
> upgraded, I have *not* upgraded any other system components (i.e.
> exactly the same environment as under 2.2.10) but have had countless
> problems. Sometimes the machine simply hangs for no apparent reason,
> processes die without explanation (logins, X, etc...) I'm used to
> Netscape crashing, but this is now at least once every 15-30 mins. gmc
> crashes once approximately every 5 mins (or after more than one or two
> file operations) and so do other file managers. Sound often fails
> completely. Machine has become more sluggish, internet connections have
> died more often than usual, mounting partitions can cause system
> crashes, and that's just the beginning. Don't even start mentioning the
> crypto stuff or anything graphical like X. :) basically what I'm saying
> is that my routine system problems have been magnified probably by an
> order of mahnitude over the last couple of weeks. I *fear* trying to run
> more than a couple of things in X because otherwise it crashes or the
> system hangs.... I guess I'll have to downgrade the kernel to 2.2.10 if
> this continues. I just hope that we have 2.2.12 soon.

     I've got two 4-CPU Sun SS-10's here that I was having a lot of spinlock() 
problems with 2.2.10, after applying a number of patches people provided, it
was getting down to only about a crash a week on average (more often on web
server, less often on news server). 

     Going to 2.2.11, with the patch that fixes the TCP memory leak installed,
it was much worse, crash maybe every 6-12 hours, but also on the news server,
innd would OOPs and die, core dump and die, or sometimes it would silently
die, no core, no oops, just no longer running.

     On the web server, 2.2.11 also generated OOP's occasionally, CPU usage
was noticably higher than with 2.2.10, and there were these weird CPU spikes
about once every 2-3 hours when the machine was busy that weren't present with
2.2.10.

     I will provide more details as time allows, but being as these machines
actually have to provide services I've got back to 2.2.10 in the interim.  (for
those of you here at eskimo, that's what reboots of www and eskinews were about
earlier this evening). 

     I have to tell you I am starting to have paranoid thoughts about someone
from M$ having planted bombs in 2.2.11.  I noticed earlier on the list that
someone mentioned problems with mmap() in 2.2.11, since both the web server and
the news server use mmap() that might be related.