On Wed, Nov 7, 2018 at 4:05 AM <arnold@skeeve.com> wrote:

Dan Cross <crossd@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.

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
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.