[comp.lang.smalltalk] Catatonia cure

rentsch@unc.UUCP (Tim Rentsch) (12/21/86)

This is my promised follow-up posting on 'curing' catatonia by an
adb patch to the unix kernel.  

First, the news.  The cure works great!  The test was done on
several suns, and done without modifying the configuration *except*
to install the patch.  The results:

	(1) longer runs of a large application, and greater user
	confidence in the solidity of the interpreter

	(2) hours of running on one particularly problematic sun,
	with no crashes (not conclusive, but much better than past
	performance)

	(3) when moving the mouse continuously during a save, things
	work fine.  This is significant because without the patch
	installed, this *repeatably* causes problems (unrepeatably
	it causes crashes, but even when there is no crash there are
	problems with queue overflow).  After the patch is
	installed, this repeatable symptom disappears.

Minor side note:  although Smalltalk was causing crashes, some
symptoms I reported after a crash may have been caused by a bad CPU
board.  You can decide for yourself what this means, I just wanted
to set the record straight.  This really is just a side issue, since
(1) Smalltalk definitely was causing crashes, and (2) those crashes
have stopped happening since we installed the patch that fixes the
sun window system bug.

And, for those intrepid folks who would like to try kernel stomping
for themselves, here is the patch:

    There was a 3.0 kernel deadlock bug that may be the cause.  The
    fix was to set the kernel variable ws_set_favor to zero.  

	% adb -w /vmunix
	  ws_set_favor?W 0
	  ^D
	% /etc/reboot.


That's all.

cheers,

txr

rentsch@unc.UUCP (Tim Rentsch) (12/23/86)

Prompted by the following:
> Hi -- I manage a bunch of suns on which we run smalltalk (rather
> heavily used) - We are currently upgrading to release 3.2 -- do you
> know if the same problem exists under 3.2?  

I realize I should have been more explicit about what versions we
are running, etc., in talking about catatonia and its cure.

The problems have been observed on both 3.0, and 3.2 PILOT.  The
patch was installed, with positive effects, in both 3.0, and in 
3.2 PILOT.  I have been told the problem will be fixed in 3.2,
presumably after PILOT (which I don't know the name of).

So, if you have 3.0 or 3.2 PILOT, I would say the patch should be
installed.  If you have 3.2 after PILOT, check and let us know.

If someone out there has a path handy for some Sun person who would
(definitively) know, you might send this message their way, to get
an answer about 3.2 after PILOT.  Followup to this newsgroup, 
*por favor*.  (How about French?  Can you say it in French? :-)

cheers,

txr