Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!wuarchive!cs.utexas.edu!rice!sun-spots-request
From: auspex!guy@uunet.uu.net (Guy Harris)
Newsgroups: comp.sys.sun
Subject: Re: SunOS 4.0 and memory
Keywords: Miscellaneous
Message-ID: <3272@brazos.Rice.edu>
Date: 18 Nov 89 20:48:20 GMT
Sender: root@rice.edu
Organization: Sun-Spots
Lines: 60
Approved: Sun-Spots@rice.edu
X-Refs:  Original: v8n192, Replies: v8n196 v8n199
X-Sun-Spots-Digest: Volume 8, Issue 206, message 7 of 15

>I have heard that 4.0 can be configured (more like stripped) to run in 4
>Mb, but it's pretty marginal.  This is still at the level of rumor, but
>perhaps a slightly more detailed rumor.

As someone who, on a regular basis, uses a diskless 3/50 with 4MB of
memory, running SunOS 4.0.3, I can contribute at least one non-rumor data
point:

I run with a configured kernel (a MUST for 4.x on small machines - but
then you're wasting memory if you run a GENERIC kernel for ANY release of
SunOS, including 3.x, 2.x, or 1.x, or probably for any other system with
BSD-flavored configuration mechanisms, unless you have one of every
device, file system, etc. that the OS supports); the configuration file is
vanilla DL50.

I have a sparse desktop (a "cmdtool" to capture console output, two or
three 54x80 shelltools, a shelltool running a mail-watcher script) and no
fancy root window backgrounds.  (I'm running the 4.0.3 versions of the
Sunview stuff, not the 3.x versions.)

I find the performance acceptable, although it could certainly be better.
I run compiles on my machine; it does have some impact on the other
"shelltool" (usually logged into a 4/280 here), and isn't blazingly fast,
but is "good enough".  I sometimes fire up Berkmail on my machine while
the compile is running; it takes a while, but then again:

	    auspex% ls -l /usr/spool/mail/guy
	    -rw-------  1 guy       2567314 Nov 18 11:55 /usr/spool/mail/guy

(I cut and pasted that; the number is NOT a typo, it really is that big).
It does fire up much faster on the 4/280, but that has 32MB of memory, a
much faster CPU, and a direct connection to the disks on which
"/usr/spool/mail" resides (no, we're not using the 4/280 as our primary
server, it just has "/usr/spool/mail" on it; the primary server is, as you
might expect, an Auspex NS5000, so we do use our own product).

This doesn't include all the assorted tuning suggestions Sun has put out.
I still run "routed", but then we don't have enough machines that the
route table fills up memory; if you have a big network, turning "routed"
off on non-multi-homed hosts may help.  I still run "sendmail", just in
case somebody sends mail to me on my home machine rather than on "auspex",
although outgoing mail from here now has internal host names stripped out
in the hopes that mail delivery doesn't involve the user's machine.

Note: if you're running only X11 (not X11/NeWS), and not Sunview, you may
want to consider removing the "pseudo-device" lines for "dtop" and "win",
as the MIT server, as I remember, gets keyboard and mouse input directly
from "/dev/kbd" and "/dev/mouse" and doesn't require all the Sunview
support kernel goo (which, I seem to remember from a while ago, takes
about 70KB on a Sun-3 - it includes pixrect code for cursor tracking).

None of this, of course, says "memory upgrades don't do any good"; I
suspect things would run better with 8MB.  It does, however, say that, at
least in some situations, 4MB is *not* unusable.  Your mileage may, of
course, differ.  (I don't run big LISP jobs, for example.  :-)  I also
don't run GNU Emacs; I run MicroEMACS at present, which "size" tells me is
smaller than "vi".)

The statements above held true before I started running 4.0.3, and was
running 4.0.1 (and, I think, back when I was running 4.0.)