6 Nov
2018
6 Nov
'18
1:41 a.m.
Spurred by the recent discussion of NIS, NIS+, LDAP et al, I'm curious what
the landscape was like for distributing administrative information in early
Unix networks.
Specifically I'm thinking about things like the Newcastle Connection, etc.
I imagine that PDP-11's connected to the ARPAnet running Unix would (e.g.,
RFC 681 style) would have adapted the HOSTS.TXT format somehow. What about
CHAOS? Newcastle? Datakit?
What was the introduction of DNS into the mix like? I can imagine that that
changed all sorts of assumptions about failure modes and the like.
NIS and playing around with Hesiod are probably the earliest such things I
ever saw, but I know there must have been prior art.
Supposedly field 5 from /etc/passwd is the GECOS username for remote job
entry (or printing)? How did that work?
- Dan C.
7 Nov
7 Nov
9:05 a.m.
Dan Cross <crossd(a)gmail.com> wrote:
What was the introduction of DNS into the mix like? I
can imagine that that
changed all sorts of assumptions about failure modes and the like.
I'm not sure what you're asking. When DNS came along, it became
a matter of editing /etc/nsswitch.conf to include dns as one of the
options along with files and yp/nis. I think the average user didn't
see any big difference since all the apps (ftp, telnet) just went
through gethostbyname().
DNS configuration files were a trip, especially for someone used to the
very simple Unix configuration files like /etc/passwd and /etc/hosts.
Circa 1985/1986 I was responsible for bringing up DNS at the Emory U
campus. Once in place, things pretty much just worked. Or at least,
from this distant vantage point, that's what I seem to remember.
Arnold
3:52 p.m.
On Wed, Nov 7, 2018 at 3:38 AM <arnold(a)skeeve.com> wrote:
I'm not sure what you're asking. When DNS
came along, it became
a matter of editing /etc/nsswitch.conf to include dns as one of the
options along with files and yp/nis.
This does not align with my memory at all. I was at udel until 1988
and we started dealing with dns ca. 1986, and the shared library stuff
I dealt with in sunos came later.
Was there really an nsswitch.conf before 1988 that used shared
libraries? When did you first see nsswitch.conf? I first saw the
shared libraries with sunos 4.0 when I moved to super.org in 1988. I
always thought it started with them on Unix anyway.
In any event, dns was in some ways a big improvement in life. We were
ftp'ing the 256KiB host file from prep frequently (along with many)
and the load on prep, evidently, was getting high.
And a linear search of a 256KiB hosts file for every single
gethostbyname was getting ... noticeable. I assume this is part of why
stayopen was an option in the library.
People got upset about some things with DNS: "it's called prep, not
prep.ai.mit.edu, what is this nonsense?"
But we adjusted.
6:27 p.m.
On 11/7/2018 10:52 AM, ron minnich wrote:
On Wed, Nov 7, 2018 at 3:38
AM<arnold(a)skeeve.com> wrote:
I first saw nsswitch.conf on Solaris. And
lo-and-behold:
"Sun Microsystems <https://en.wikipedia.org/wiki/Sun_Microsystems> first
developed NSS for their Solaris
<https://en.wikipedia.org/wiki/Solaris_%28operating_system%29> operating
system, but subsequently programmers ported it to many other operating
systems including FreeBSD <https://en.wikipedia.org/wiki/FreeBSD>,
NetBSD <https://en.wikipedia.org/wiki/NetBSD>, Linux
<https://en.wikipedia.org/wiki/Linux>, HP-UX
<https://en.wikipedia.org/wiki/HP-UX>, IRIX
<https://en.wikipedia.org/wiki/IRIX> and AIX
<https://en.wikipedia.org/wiki/AIX_operating_system>."
https://en.wikipedia.org/wiki/Name_Service_Switch
How true that is, I'd love to know.
art k.
I'm not sure what you're asking. When
DNS came along, it became
a matter of editing /etc/nsswitch.conf to include dns as one of the
options along with files and yp/nis.
This does not align with my memory at all. I
was at udel until 1988
and we started dealing with dns ca. 1986, and the shared library stuff
I dealt with in sunos came later. 
9:28 p.m.
I remember that early linux used /etc/host.conf -- or hosts.conf for DNS vs hosts file or
to.
Hacking the SunOS4.1.3 library with the resolv21 fixes to add DNS w/o NIS was fun too.
Both about 1991 for me iirc.
Bill
Sent from MailDroid
-----Original Message-----
From: Arthur Krewat <krewat(a)kilonet.net>
To: tuhs(a)minnie.tuhs.org
Sent: Wed, 07 Nov 2018 16:22
Subject: Re: [TUHS] Directory services in early Unix networks?
On 11/7/2018 10:52 AM, ron minnich wrote:
On Wed, Nov 7, 2018 at 3:38
AM<arnold(a)skeeve.com> wrote:
I first saw nsswitch.conf on Solaris. And
lo-and-behold:
"Sun Microsystems <https://en.wikipedia.org/wiki/Sun_Microsystems> first
developed NSS for their Solaris
<https://en.wikipedia.org/wiki/Solaris_%28operating_system%29> operating
system, but subsequently programmers ported it to many other operating
systems including FreeBSD <https://en.wikipedia.org/wiki/FreeBSD>,
NetBSD <https://en.wikipedia.org/wiki/NetBSD>, Linux
<https://en.wikipedia.org/wiki/Linux>, HP-UX
<https://en.wikipedia.org/wiki/HP-UX>, IRIX
<https://en.wikipedia.org/wiki/IRIX> and AIX
<https://en.wikipedia.org/wiki/AIX_operating_system>."
https://en.wikipedia.org/wiki/Name_Service_Switch
How true that is, I'd love to know.
art k.
I'm not sure what you're asking. When
DNS came along, it became
a matter of editing /etc/nsswitch.conf to include dns as one of the
options along with files and yp/nis.
This does not align with my memory at all. I
was at udel until 1988
and we started dealing with dns ca. 1986, and the shared library stuff
I dealt with in sunos came later.
5:15 p.m.
On Wed, Nov 7, 2018 at 4:05 AM <arnold(a)skeeve.com> wrote:
Dan Cross <crossd(a)gmail.com> wrote:
Sorry, this was indeed vague.
Ron alluded to what I was asking about; namely, what were the circumstances
that gave rise to the creation of DNS in the first place? I imagine an
unwieldy HOSTS.TXT file being FTP'd daily combined with a linear scan (or
an expensive step to create a DBM file or something) had a lot to do with
it.
When DNS came along, it became
What was the introduction of DNS into the mix
like? I can imagine that
that
changed all sorts of assumptions about failure
modes and the like.
I'm not sure what you're asking. a matter of editing /etc/nsswitch.conf to include dns
as one of the
options along with files and yp/nis. I think the average user didn't
see any big difference since all the apps (ftp, telnet) just went
through gethostbyname().
This doesn't mesh with my memory. I recall building BIND from source and
having to rebuild network programs (e.g. on 4.3 on the RT or VAXen) to pick
up the new version of libresolv.a, and hacking the resolver library into
libc.so on Suns. I remember using resolv.conf fairly early on, but my
memory is that nsswitch.conf came later (Solaris 2.x era?). Ultrix did have
a configuration file for where to do host lookups, but I think the set of
sources was fixed: files, NIS or DNS. This would have been in the Ultrix
4.4 or 4.5 era on MIPS. I remember seeing some description of a
configuration file accompanied by an editorialized comment saying something
like, "this is an idea that's time has come: Ultrix has had it for several
years." The dig on uglix was, well, kind of funny (I had a DECstation at
home at the time).
DNS configuration files were a trip, especially for someone used to the
very simple Unix configuration files like /etc/passwd
and /etc/hosts.
Circa 1985/1986 I was responsible for bringing up DNS at the Emory U
campus. Once in place, things pretty much just worked. Or at least,
from this distant vantage point, that's what I seem to remember.
Ha! Yeah, tell me about it. We used to format our SOA records so that the
'(' was on a line by itself, indented to 'SOA'. At some point, a
minor
point revision to BIND changed the parsing behavior so that this was a
syntax error; there was no warning that this would be the case. I
complained on a BIND mailing list and was told, "this is in the RFC." It
wasn't clear to me where, in the documentation, it was specified that the
file format understood by BIND was the RFC-standard format.
- Dan C.
8:15 p.m.
On Wed, Nov 07, 2018 at 12:15:32PM -0500, Dan Cross wrote:
On Wed, Nov 7, 2018 at 4:05 AM
<arnold(a)skeeve.com> wrote:
When DNS came along, it became
Yeah, this is right, I remember doing this as well and I worked at Sun
at that point. It was a pain.
a matter of editing /etc/nsswitch.conf to include
dns as one of the
options along with files and yp/nis. I think the average user didn't
see any big difference since all the apps (ftp, telnet) just went
through gethostbyname().
This doesn't mesh with my memory. I recall building BIND from source and
having to rebuild network programs (e.g. on 4.3 on the RT or VAXen) to pick
up the new version of libresolv.a, and hacking the resolver library into
libc.so on Suns. I remember using resolv.conf fairly early on, but my
memory is that nsswitch.conf came later (Solaris 2.x era?).
8 Nov
8 Nov
11:34 a.m.
Larry McVoy <lm(a)mcvoy.com> wrote:
On Wed, Nov 07, 2018 at 12:15:32PM -0500, Dan Cross
wrote:
OK, so now that everyone mentions it, I was wrong. I think I only
had to do this same kind of thing once though.
We ran the Mt. Xinu 4.3 BSD + NFS and it had YP also, so at some
point they were doing it and I sorta think we had /etc/nsswitch.conf
on the vaxen, but I could be wrong. We definitely had it on the SunOS 4.0
Sun servers that we bought later on.
Our connection was via CSNet - slow X.25 line to Georgia Tech and from
there to the rest of the world. We most certainly did NOT do daily FTPs
of the HOSTS.TXT file or anything like that. At the time we had very
few TCP/IP-savy users; you could probably count them on two hands and
have a few fingers left over. :-)
Those were the days. :-)
Arnold
On Wed, Nov 7, 2018 at 4:05 AM
<arnold(a)skeeve.com> wrote:
When DNS came along, it became
Yeah, this is right, I remember doing this as well and I worked at Sun
at that point. It was a pain. a matter of editing /etc/nsswitch.conf to include
dns as one of the
options along with files and yp/nis. I think the average user didn't
see any big difference since all the apps (ftp, telnet) just went
through gethostbyname().
This doesn't mesh with my memory. I recall building BIND from source and
having to rebuild network programs (e.g. on 4.3 on the RT or VAXen) to pick
up the new version of libresolv.a, and hacking the resolver library into
libc.so on Suns. I remember using resolv.conf fairly early on, but my
memory is that nsswitch.conf came later (Solaris 2.x era?). 
11:14 p.m.
On Thu, Nov 8, 2018, 1:07 PM Arthur Krewat <krewat(a)kilonet.net wrote:
On 11/8/2018 6:34 AM, arnold(a)skeeve.com wrote:
Early directory services were rcp/ftp, in all their glory. :). Mostly
though at the small school I was at it was moot. THE VAX we had was 100%
ASCII terminals to access it, except if you were logging in from the
Tops-20 box via telnet from one of its ASCII terminals.
SunOS 4 definitely had YP. We used it in school when the Tops-20 machine
was replaced by a boatload of 68000 machines running SunOS and Suntools...
I can't recall if 3.5 also had it or not... it was later it changed to NIS.
Warner
We definitely had it on the SunOS 4.0
Sun servers that we bought later on.
Still incorrect. ;)
It first showed up in Solaris, not SunOS.
9 Nov
9 Nov
12:06 a.m.
On 11/8/2018 6:14 PM, Warner Losh wrote:
SunOS 4 definitely had YP. We used it in school when the Tops-20
machine was replaced by a boatload of 68000 machines running SunOS and
Suntools... I can't recall if 3.5 also had it or not... it was later
it changed to NIS.
Anything that had Sun's NFS had YP, at least as of version 2.0. It was
part of the package.
Somewhere in Warren's archive is a source copy of NFS 2.0 from Sun that
I donated. It's replete with all the YP stuff. Not sure if the following
is the copy I donated, but it's close enough:
https://minnie.tuhs.org//cgi-bin/utree.pl?file=NFSv2/usr/src/usr.bin/ypcat.c
From my copy:
-r--r--r-- 1 krewat kilonet 6356 Nov 20 1985
./usr/src/usr.bin/ypcat.c
Interesting that the date is one day off.
The copy of NFS 2.0 that I donated was integrated with BSD 4.2 at
Computer Graphics Lab at New York Institute of Technology. When they
went kaput, I managed to get my hands on that NFS tape along with a slew
of other stuff that I can't do anything with because it's all copywrited.
/etc/nsswitch.conf in Solaris originated with NIS+ (NISPLUS). Sun seems
to be the origin, and it only came about because of NIS+. That's my
theory and I'm sticking with it ;)
art k.
7 Nov
7 Nov
8:15 p.m.
On Wed, 7 Nov 2018 at 14:57, Dan Cross <crossd(a)gmail.com> wrote:
This doesn't mesh with my memory. I recall building BIND from source and
having to rebuild network programs (e.g. on 4.3 on the RT or VAXen) to pick
up the new version of libresolv.a, and hacking the resolver library into
libc.so on Suns. I remember using resolv.conf fairly early on, but my
memory is that nsswitch.conf came later (Solaris 2.x era?). Ultrix did have
a configuration file for where to do host lookups, but I think the set of
sources was fixed: files, NIS or DNS. This would have been in the Ultrix
4.4 or 4.5 era on MIPS. I remember seeing some description of a
configuration file accompanied by an editorialized comment saying something
like, "this is an idea that's time has come: Ultrix has had it for several
years." The dig on uglix was, well, kind of funny (I had a DECstation at
home at the time).
Ultrix 4.0 (1990) had /etc/svc.conf for controlling distributed service
lookups, and you are correct that the only options were local, yp, and
bind. Ultrix 3 (1988) had /etc/svcorder which was much more limited, only
allowing for setting the order of host lookups, but it did have support for
resolv.conf and BIND lookups (which still works!).
-Henry
9:11 p.m.
On Wed, Nov 7, 2018 at 9:57 PM Dan Cross <crossd(a)gmail.com> wrote:
On Wed, Nov 7, 2018 at 4:05 AM
<arnold(a)skeeve.com> wrote:
I always assumed that nsswitch came much later than DNS, based just on the
fact that Linux glibc on older systems still ships an /etc/host.conf
"resolver configuration file" that has such parameters as "order
hosts,bind,nis" (although they're inoperative).
--
Mantas Mikulėnas
When DNS came along, it became
a matter of editing /etc/nsswitch.conf to include dns as one of the
options along with files and yp/nis. I think the average user didn't
see any big difference since all the apps (ftp, telnet) just went
through gethostbyname().
This doesn't mesh with my memory. I recall building BIND from source and
having to rebuild network programs (e.g. on 4.3 on the RT or VAXen) to pick
up the new version of libresolv.a, and hacking the resolver library into
libc.so on Suns. I remember using resolv.conf fairly early on, but my
memory is that nsswitch.conf came later (Solaris 2.x era?). Ultrix did have
a configuration file for where to do host lookups, but I think the set of
sources was fixed: files, NIS or DNS.
3:02 p.m.
On Mon, Nov 5, 2018 at 7:35 PM Dan Cross <crossd(a)gmail.com> wrote:
Spurred by the recent discussion of NIS, NIS+, LDAP et
al, I'm curious
what the landscape was like for distributing administrative information in
early Unix networks.
Specifically I'm thinking about things like the Newcastle Connection, etc.
I imagine that PDP-11's connected to the ARPAnet running Unix would (e.g.,
RFC 681 style) would have adapted the HOSTS.TXT format somehow. What about
CHAOS? Newcastle? Datakit?
What was the introduction of DNS into the mix like? I can imagine that
that changed all sorts of assumptions about failure modes and the like.
NIS and playing around with Hesiod are probably the earliest such things I
ever saw, but I know there must have been prior art.
Supposedly field 5 from /etc/passwd is the GECOS username for remote job
entry (or printing)? How did that work?
Dan - all good questions, but I think you are mixing a few things (which is
easy to do as they all had different evolutionary paths).
- ARPAnet was Rand, UCLA and UofI in the early to mid 70s.
- UCLA line would fork competely with the original Locus work of the mid
70's, which would reappear later in the 80's post BSD
- IP Networking was done by BBN for 4.1BSD in the late 70s - originally
as an OS independant stack (hence it has its own memory manager to
insulated it from the local S). Besides UNIX I think it went into HP's MPE
and maybe a couple f other systems.
- The BBN IP stack was then repliced into UNIX by UCB/CRSG as 4.1A with
Joy's sockets layer in 82/83
- HOST.TXT was finaly abandoned and BIND was then done (primarily at UCB
by peed on by many) - I want to say eary 80's the SCCS files might give
you some hints.
- Hesiod was MIT/Athenia and NIS by Sun were later developed somewhat
in the same time frame (mid to late 80s)
- CHAOS was completely seperate, although influenced the BBN code and
was the early/mid 70s.
- BTL's DataKit of course, had the UoI (Chesson) influence was late 70s.
- Best I can tell Newcastle was complete seperate from all of this (also
late 70s).
Clem
ᐧ
3:05 p.m.
And I left out the whole X.500 DAP which begat ??UMich's I think? LDAP -
which was mid to late 1980s.
ᐧ
On Wed, Nov 7, 2018 at 10:02 AM Clem Cole <clemc(a)ccc.com> wrote:
On Mon, Nov 5, 2018 at 7:35 PM Dan Cross <crossd(a)gmail.com> wrote:
Spurred by the recent discussion of NIS, NIS+,
LDAP et al, I'm curious
what the landscape was like for distributing administrative information in
early Unix networks.
Specifically I'm thinking about things like the Newcastle Connection, etc.
I imagine that PDP-11's connected to the ARPAnet running Unix would
(e.g., RFC 681 style) would have adapted the HOSTS.TXT format somehow. What
about CHAOS? Newcastle? Datakit?
What was the introduction of DNS into the mix like? I can imagine that
that changed all sorts of assumptions about failure modes and the like.
NIS and playing around with Hesiod are probably the earliest such things
I ever saw, but I know there must have been prior art.
Supposedly field 5 from /etc/passwd is the GECOS username for remote job
entry (or printing)? How did that work?
Dan - all good questions, but I think you are mixing a few things (which
is easy to do as they all had different evolutionary paths).
- ARPAnet was Rand, UCLA and UofI in the early to mid 70s.
- UCLA line would fork competely with the original Locus work of the
mid 70's, which would reappear later in the 80's post BSD
- IP Networking was done by BBN for 4.1BSD in the late 70s -
originally as an OS independant stack (hence it has its own memory manager
to insulated it from the local S). Besides UNIX I think it went into HP's
MPE and maybe a couple f other systems.
- The BBN IP stack was then repliced into UNIX by UCB/CRSG as 4.1A
with Joy's sockets layer in 82/83
- HOST.TXT was finaly abandoned and BIND was then done (primarily at
UCB by peed on by many) - I want to say eary 80's the SCCS files might
give you some hints.
- Hesiod was MIT/Athenia and NIS by Sun were later developed somewhat
in the same time frame (mid to late 80s)
- CHAOS was completely seperate, although influenced the BBN code and
was the early/mid 70s.
- BTL's DataKit of course, had the UoI (Chesson) influence was late
70s.
- Best I can tell Newcastle was complete seperate from all of this
(also late 70s).
Clem
ᐧ
5:02 p.m.
On 11/7/2018 7:02 AM, Clem Cole wrote:
* ARPAnet was Rand, UCLA and UofI in the early to
mid 70s.
For what it's worth, the first 4 nodes on the Arpanet were
UCLA, SRI, UCSB (my alma mater), and U of Utah.
I arrived at UCSB in 1973 and its presence on the ARPAnet wasn't
something that most people there knew about. I certainly didn't.
Jon
7:08 p.m.
Dan - all good questions, but I think you are mixing a
few things (which is
easy to do as they all had different evolutionary paths).
- ARPAnet was Rand, UCLA and UofI in the early to mid 70s.
- UCLA line would fork competely with the original Locus work of the mid
70's, which would reappear later in the 80's post BSD
- IP Networking was done by BBN for 4.1BSD in the late 70s - originally
as an OS independant stack (hence it has its own memory manager to
insulated it from the local S). Besides UNIX I think it went into HP's MPE
and maybe a couple f other systems.
- The BBN IP stack was then repliced into UNIX by UCB/CRSG as 4.1A with
Joy's sockets layer in 82/83
- HOST.TXT was finaly abandoned and BIND was then done (primarily at UCB
by peed on by many) - I want to say eary 80's the SCCS files might give
you some hints.
- Hesiod was MIT/Athenia and NIS by Sun were later developed somewhat
in the same time frame (mid to late 80s)
- CHAOS was completely seperate, although influenced the BBN code and
was the early/mid 70s.
- BTL's DataKit of course, had the UoI (Chesson) influence was late 70s.
- Best I can tell Newcastle was complete seperate from all of this (also
late 70s).
Clem
ᐧ
Perhaps also worthy of a mention is Network Registration Service (NRS)
used in the UK academic network (JANET). Similar idea to
HOSTS.TXT. Periodic download and munge.
More interestingly though, NRS used the opposite ordering to DNS, so
"cs.nott.ac.uk" was "uk.ac.nott.cs", which became a problem when
Czechoslovakia got their TLD (.cs) in the 80s as the both DNS and NRS
were used in parallel for some period of time.
I believe NRS was still being used to some capacity by the time
Czechoslovakia split in 93.
Aaron
--
Aaron Jackson - M6PIU
http://aaronsplace.co.uk/
7:48 p.m.
On Wed, Nov 7, 2018 at 9:43 AM Clem Cole <clemc(a)ccc.com> wrote:
ARPAnet was Rand, UCLA and UofI in the early to mid
70s.
And my alma mater, the University of Utah
(https://www.lib.utah.edu/digital-scholarship/arpanet/index.php)
--
Jim
7:51 p.m.
Sorry - if this was not clear -- the original UNIX work for the ArpaNet
was UCL, Rand and UoI
ᐧ
On Wed, Nov 7, 2018 at 2:48 PM Jim Davis <jim.epost(a)gmail.com> wrote:
On Wed, Nov 7, 2018 at 9:43 AM Clem Cole
<clemc(a)ccc.com> wrote:
ARPAnet was Rand, UCLA and UofI in the early to
mid 70s.
And my alma mater, the University of Utah
(https://www.lib.utah.edu/digital-scholarship/arpanet/index.php)
--
Jim
1988
days inactive
1991
days old
18 comments
13 participants
participants (13)
-
Aaron Jackson -
arnold＠skeeve.com -
Arthur Krewat -
Clem Cole -
Dan Cross -
Henry Bent -
Jim Davis -
Jon Forrest -
Larry McVoy -
Mantas Mikulėnas -
ron minnich -
Warner Losh -
William Pechter