Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cornell!uw-beaver!rice!sun-spots-request
From: galvin-peter@yale-bulldog.arpa (Peter Baer Galvin)
Newsgroups: comp.sys.sun
Subject: Re: Diskless Client under 4.0
Keywords: SunOS
Message-ID: <52141@yale-celray.yale.UUCP>
Date: 8 Mar 89 14:24:31 GMT
References: <8902062031.AA03008@macduff.ultra.com> <2634@eos.UUCP>
Sender: usenet@rice.edu
Organization: Yale University Computer Science Dept, New Haven CT  06520-2158
Lines: 32
Approved: Sun-Spots@rice.edu
Original-Date: 28 Feb 89 16:35:34 GMT
X-Sun-Spots-Digest: Volume 7, Issue 187, message 3 of 15

This brings up something I've wanted to mention for a while but never got
around to.  Under 4.0 you are now able to create many separate swap files
for use by one client, allowing easy addition of swap space as needed.
Isn't that convenient...except that there is a noticible performance
penalty.

A user had 16mb of swap and needed another 8, so I created (using mkfile)
another 8mb file on a separate partition on the same disk and added it to
the workstation via swapon.  The user came back saying that the system was
now _much_ slower than it had been...maybe by 1/3.  I called this in to
sun, after more experimentation, and asked about it at the SUG conference
too.  The best guess was that we were being hit by disk head seeks, and
that unless we had two disks on two controllers to do separate seeks and
transfers, (swap files on each of the two disks) we would see this kind of
problem.

I then built a 24mb swap file on the other partition, and the problem went
away, so the guess seems correct.  I'm a little surprised though, since in
the old days we had 0b and 1b partitions on separate disks on the same
controller on our servers and didn't notice and performance problems.
You'd think that separate swap files on two disks on one controller would
perform similarly, but I tried this configuration also, on a workstation,
and saw the performance problem again.  Of course we're using the usual
dumb Sun disk controller, so just changing that for a 7053 (or whatever)
might help.

--Peter

Peter Baer Galvin       		      (203)432-1254
Senior Systems Programmer, Yale Univ. C.S.    galvin-peter@cs.yale.edu
51 Prospect St, P.O.Box 2158, Yale Station    ucbvax!decvax!yale!galvin-peter
New Haven, Ct   06457			      galvin-peter@yalecs.bitnet