Subject: Re: Public Testdrive: NetBSD 1.5.2/Alpha
To: David Brownlee <>
From: Stephen M Jones <>
List: netbsd-users
Date: 12/28/2001 13:57:33
David wrote:

> 	Would you be interested in belign listed in the 'sites running
> 	NetBSD' page - maybe even on the testimonials? We'd be more than
> 	willing to include a link with details on how to sign up.

I certainly would have no problem with that.

I actually started using NetBSD for test & production environments at (a private network of hosts).  I administered to
a number of machines running NetBSD-1.2.1.  These machines ranged from x86
to sparc and hp300.  Eric Schnoebelen who I believe was involved in the
hp300 port for NetBSD was the software engineer I was supporting.  I
also had my own x86 machine on the network which started out running
ESIX/ISC interactive, then for a very brief time linux and ever since
then (1996) it has run NetBSD.  All of these systems have responsibilities
ranging from USENET news spooling, uucp, smtp, pop3, httpd, anon ftpd
user shell access, et cetera

Prior to this ..

My father who worked for Western Electric and Bell-labs from the early
1960's to 1979 introduced me to my first teletype at the age
of 6. We used it along with a 300 baud modem w/ acoustic coupler to
dial up into a Bell Labs UNIX system so he could do work and then later
we'd play games.  But honestly, we mostly played games.  

(insert more info .. timex sinclair 1000, vic-20, c64, apple .. though
I am proud to say the first computer I ever used was a DEC PDP-11)

In 1987 a friend of mine started an electronic bulletin board called 'SDF-1'
running on an Apple ][e.  I pitched in and started helping out.  We had
also been using a PUBNIX in Dallas Texas called 'killer' or 'attctc' which
was run by Charlie Boykin on an AT&T 3B2.  In 1990, some of you may remember,
killer, including a few others (Steve Jackson games, Jolinet) were shutdown by
the Secret Service during 'operation sundevil'.  We decided to form a
replacement system for killer and thus the 'SDF public access UNIX system'
was spawned from the SDF-1.

The time line looks roughly a bit like this:

1987-1990 Apple ][ ProDOS
1989-1990 Co-Herent  (incoherent?) x86
1990-1993 ISC SystemV, ESIX x86
1993-1997 AT&T SystemV, 3B2/522 & 3B2/1000
1997-2001 Linux x86

As mentioned before, a private machine of mine began runing NetBSD 1.2.1
in 1996.  At the same time, I was looking to move on from the 3b2's now that
PC hardware was getting cheaper and faster.  Unfortunately this was the
worst decision for me (the move back to x86).  I attempted to use NetBSD,
but I was supporting over 10,000 users at the time.  I got frustrated
with how pwd_mkdb worked.  I realise this problem still exists, but I
have a work around (I'll write a little more about that in just a bit).
This was critical, because the password file needs to be updated frequently
each day and quickly.  I gave up and went with linux (I also attempted
using GNU hurd).  At first it seemed just fine.  Linux was still not 
widely known but buffer overruns were just getting their start.  I had alot
of pressure from people who were convinced that linux was the holy grail
and that running SystemV on the 3b2 was a joke.  So, I ran linux.  It 
was ok, but just *ok* .. the real problems of course lay ahead ranging
from flaws in the ext2 file system, the crappiness of consumer computer
hardware and the crufty lossage that is the linux kernel.  I won't go into
details as its just to horrible to recall that waste of time.  I'll call it
the 'sdf darkages'.

I was sort of becoming apathetic towards the whole thing, uptime goals 
weren't cool anymore, if I could keep them running for 15 days, 
that was a miracle!  The 3b2s ran for 100+ days and were only taken down
because I wanted to add new hardware.  I spent a number of months out of
town, rather out of the country.  And came back with a load of money and
a desire to rescue sdf.  Earlier this year I was able to cash in on the crash (the wave that I got to see form, crest and crash) and scored
on a few DEC machines.  This time when I ran into the pwd_mkdb issue, I
got the source and had it point its pwd.db and spwd.db in a memory file
system which is generated at system boot when the mfs is formed.  That,
and slight modifications to the hashinfo parameters got me generating 
the password database in less time than it took linux's shadow suite to
modify its flat text files.  

So now this is a new chapter for the public access UNIX system, a new
decade .. a new century ;-) 

For christmas I got a coffee mug from one of the sdf users.  The picture
had 'tux' in a compromising position with the daemon taking advantage of
him.  I got this, because I had become quite hostile towards linux (and
I'm a pretty easy going guy!)  If you want to get a few chuckles:

Fri Dec 28 13:57:11 CST 2001 Details:

23651 active accounts (I purge after 90 days of inactivity)
100+ concurrent users (maxusers of 210 set on Mon Dec 3 17:41:00 UTC 2001)
250-400 new users each day
10-15 expired accounts each day
50000 successful http requests on average per day
2017 virtual hosted sub-domains (each user gets
     if they so desire).
29357 sendmail log entries since midnight today 

Proudly running NetBSD on the Dec Alpha