Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!husc6!necntc!ames!aurora!labrea!jade!eris!mwm
From: mwm@eris.BERKELEY.EDU (Mike (My watch has windows) Meyer)
Newsgroups: comp.arch
Subject: micros vs. minis for timesharing
Message-ID: <4963@jade.BERKELEY.EDU>
Date: Wed, 2-Sep-87 09:31:06 EDT
Article-I.D.: jade.4963
Posted: Wed Sep  2 09:31:06 1987
Date-Received: Sat, 5-Sep-87 10:18:50 EDT
References: <622@winchester.UUCP> <12953@amdahl.amdahl.com> <630@winchester.UUCP> <4949@jade.BERKELEY.EDU> <13439@amdahl.amdahl.com>
Sender: usenet@jade.BERKELEY.EDU
Reply-To: mwm@eris.BERKELEY.EDU (Mike (My watch has windows) Meyer)
Organization: Missionaria Phonibalonica
Lines: 63


In article <4949@jade.BERKELEY.EDU>, I wrote:
> Of course, as someone who deals with large timeshareing Unix boxes,
> I'll be happy to tell you why a Sun 3/xx at 2 or 3 11/780 mips won't
> replace a real 780..... 

So, in article <13439@amdahl.amdahl.com> mat@amdahl.amdahl.com (Mike
Taylor) writes:
<I think that would be an interesting discussion.  I, for one, would be
<interested in your views.  Perhaps you might kick things off....

Wouldn't you know I'd get called on that one. Oh well, here we go.

Two things to note here. First is "large timesharing Unix boxes," the
key word is "timesharing." The key question isn't "how fast does it
go," but "how many users can I put on it before they start to
complain."

Second is that I said 3/xx instead of 3/xxx *on purpose*. The 3/xx
boxes cover the 3/50, 3/75 and (now) the 3/60. The 3/50 is closer to a
785 than to a 780 in raw cpu (I can dig out some benchmark numbers if
people really want to see them). The 3/60 and 3/75 are both faster
than the 3/50. Well, at least they *ought* to be.

I'm not familiar with the 3/75, but the other two both suffer from the
same problem if you want to use them as timesharing systems. Lack of
I/O bandwidth. The problems I'm used to seeing on VAXen (750 and up,
smaller boxes need not apply) is that they bottleneck on something
*other* than cpu.  Usually the bottleneck is of such a nature that
putting in more I/O bandwidth will help - either directly, or by
providing faster paging.

From watching a 3/50 with people on the ethernet, 5 to 10 people is
about as many as you'd want to put on. One person using a window
manager is more than I want to put on :-).

On the other hand, with the 3/[12][68]0 machines, you can get enough
I/O bandwidth to do real timesharing. I've never tried this, but the
impression I get from talking to people who have is that the 3/1xx
machines are somewhere between a 750 and a 780 (i.e. - between 15 and
25 users before the world goes to pot), and the 3/2xx machines are
slightly better than a 780 (somewhere between 25 and 35 users).

Of course, DEC isn't selling 780s anymore, and is peddling a whole new
bus structure they claim is loads and loads faster than the old
mass/uni-bus structures.

I'd be interested in hearing from people with real experience with
using the Sun 3/[12]xx machines for timesharing, or from people who
are using the smaller BI bus machines (all I gots are 8800s, and I
ain't impressed). People with multi-micro-processor boxes are also
asked to contribute.

Arguments that "timesharing is dead, everyone is going to have a cpu
faster than a 780 on their desk soon" will be placed in the "if only
it were so" category, and ignored.

	<mike
--
How many times do you have to fall			Mike Meyer
While people stand there gawking?			mwm@berkeley.edu
How many times do you have to fall			ucbvax!mwm
Before you end up walking?				mwm@ucbjade.BITNET