[ On Sat, July 25, 1998 at 11:01:55 (-0400), Carl Shapiro wrote: ]
> Subject: Re: emacs-18.59
>
> Because emacs-20 is intolerably slow on my Sun3's with < 16MB of RAM
> (and I don't expect it to get any better once I do figure out how to
> jumper these 32MB VME RAM boards).

emacs-18 isn't much faster than emacs-19, and emacs-20 has many bugs
fixed, plus some new features that I'll never want to do without now.

Having more than 16MB or RAM definitely helps things out though,
esp. under NetBSD (as opposed to SunOS).

Setting up the 32MB boards is trivial -- exactly the same as the 16MB
and 8MB boards.  Make sure there's a terminator *ONLY* in the one
*FARTHEST* from the CPU and that the board numbers are all unique.  I
don't think board order is important, but I don't remember off hand.  In
general the Field Engineering Manual recommends the first board be in
slot 6 and the 2nd-4th be in slots 3-5.  Good power and cooling are
critically important, of course.

Note that I've had no end of trouble with a system running four 32MB
boards in a 3/280 under SunOS-4 -- there were many strange core dumps
and unexplainable system crashes.  This configuration was never
officially supported, though the 32MB boards are physically identical to
the 16MB boards that are(were) supported (save for the amount of RAM
populated on them).  Normally though the system would run A-OK for up to
a week or more.  The crashes were very disconserting though since they
seemed to corrupt random things as if the memory went totally bogus
despite the supposed ECC protection.  (I never saw many legit ECC
triggered errors, though there was the odd one that seemed to be more
likely a random glitch.)  My bet is the VME backplanes can't really
handle 4x32MB boards reliably.  (I had my ancient oscilloscope watching
some of the bus lines for a while, but I never caught anything that
looked abnormal.)

I've still not had the time or incentive though to fire that machine up
again (since I've switched to sparc) and try NetBSD on it.  My worry is
that there's not enough support in memerr.c for any of the Sun
architectures though....  I've got some knowledge of how to fix this,
though based on the header files in SunOS-4, and I've also got a copy of
some stuff Chuck Cranor and Jason Thorpe were working on to get ECC
support into NetBSD/sparc (in particular for the same VME boards it
seems).

Maybe I'll try it again after we settle into our new house (after the
end of August).  We've got a 200AMP service there, and winter will be
here sooner than we want!   ;-)

-- 
							Greg A. Woods

+1 416 443-1734      VE3TCP      <gwoods@acm.org>      <robohack!woods>
Planix, Inc. <woods@planix.com>; Secrets of the Weird <woods@weird.com>