Subject: Re: ballpark ratio users:cpu for a [345]86 NetBSD box?
To: Dave McGuire <mcguire@rocinante.digex.net>
From: Perry E. Metzger <perry@imsi.com>
List: current-users
Date: 09/27/1994 09:10:47
Dave McGuire says:
>   That would most certainly be more reasonable...But the thing is...if
> you're gonna spend a buttload of money on new hardware, why not buy
> hardware that's designed to support lots of users in the first place?
> PCs are designed to give *great* performance to one person.  Big
> servers are designed to give *good* performance to a LOT of people.
[...]
> I can pick up a Sun4/470
> for about US$2,000.00 right now, that I know from experience (I work
> for an Internet provider that sells shell accounts on Sun4/400-series
> systems) can readily handle 60 people with no problem.

Pardon, but a Pentium with a PCI bus has far better context switching,
far better IO, and far better speed. What is left that the Sun 4/470,
which was designed as a server, not as a multiuser machine (which is
why it has too few MMU contexts) can beat the Pentium machine on?

I can also pick one of those up in a reasonable configuration for
$2,000.

Its not quite the same O/S or application, but I'll note that
Compuserve has been successfully experimenting with hundreds of users
on single PCs running BSDI.

The main problem I could see with a NetBSD Pentium box right now is
that NetBSD won't max out the disk because the clustering stuff isn't
in yet. However, the hardware is as capable as anything you can
find. The Pentium is an UGLY chip in terms of instruction set, but the
notion that the machine can't compete is misplaced. Try a benchmark
before claiming otherwise.

Perry