RE: Machine naming convention

[Paul Caskey <pcaskey@swcp.com>]

On Thu, 6 Aug 1998 13:48:18 -0700 (PDT),
owner-sage-members-digest@usenix.ORG said:

Xev> From: "Gittler, Xev" <Xev.Gittler@gs.com>
Xev> Date: Thu, 6 Aug 1998 09:58:07 -0400
Xev> Subject: RE: Machine naming convention

Xev> There are essentially two camps on this issue. On one side are the
Xev> people who want everything to look like prt02nycd4, which means
Xev> something to them.  On the other side are people who would like
Xev> people readable names. There are reasonable arguments to be made
Xev> for both.

That sounds like a pretty good attitude, to me.  I waffle on this issue
because nothing seems ideal in all cases.  Ce la vie, I guess.

I started to write a lengthy justification for this "sass1752" scheme my
company has, but then I paused for a sanity check: was I really spending
time defending this crazy naming scheme?  Nah.  I've always preferred
names like bedrock, bumstead, and butane.  It drives me crazy bouncing
between three servers I work with: sahp1352, sahp1353, and sahp1354.
But that's why I use their aliases: analysis, monitor, and tooldev.

I guess I like having it both ways.  Aliases are wonderful things.  I
recently had to swap (physically and functionally) sahp1352 and
sahp1353, so I just swapped the analysis/monitor aliases in DNS and life
went on.  The "sahp1352" name has the advantage that it works in a flat
namespace across a large enterprise, and it gives some clue to the
admins as to what it is.  The aliases have the advantage that they can
be easy to remember, and can change as the machine's function, owner, or
location changes.

Using both seems to satisfy all cases.  If you try to find one name to
fit both purposes, you're absolutely stuck with a "hydra, jupiter, jets"
name.  You can't make it more descriptive in any way, because it will
either be too hard to remember and type, or it will be something like
"janespc" which will haunt you forever when Jane quits her day job and
becomes a nightclub singer.

I totally agree you should never encode something like the OS type, OS
version, IP address, room number, or user name in the hostname.  Your
examples of "etsmns052" and "nbtodc032" are horrific.  My sole defense
of the silly "sahp1352" is that there are only two things encoded: the
site, and the vendor.  The machine could move to another site, but in
such a case it will usually be scrubbed and renamed, anyway.  I guess I
think site is acceptable, but room number is not.  Debatable.

But the vendor of a machine is something that will NEVER change, and it
has the added benefit of implicitly telling you the TYPE of hardware,
because Cisco doesn't make PC's (yet).  I've found this very useful.


Xev> One scheme that worked well is naming groups of
Xev> machines with similar names, so that any football team, for
Xev> instance, represent one area. When an SA sees a machine named
Xev> 'jets' he knows that it in the 'football' group (or the 'planes'
Xev> group).  This tends to work in medium sized companies best.
Xev>          ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

That's the one thing that got me thinking that machine-generated names
were not pure evil: Cute names don't scale.  I would argue that RFC 1178
(dated Aug 1990) is out of date in this regard.  It does not address
modern issues of sane naming across a large enterprise.  

The only way to use friendly names across thousands of machines is to
divide them into DNS subdomains, which is great, except you're screwed
if "frodo" moves from this.sandia.gov to that.sandia.gov and there's
already a "frodo.that.sandia.gov".  Maybe this is a rare case, but as
companies get larger, people and machines move around more, and switched
VLANs become popular, it seems relevant.  You want your name to stay
static just like your IP address, for everyone's sanity.

In fact, even in medium-sized company, cute names don't work when people
move, because you get the "goedel, escher, fluffy" syndrome.  Your naming
schemes go down the toilet.

One last thought: the machine-generated names work REALLY well for us on
networking equipment.  If we had "betty, wilma, fred" names for all of
our routers, switches, and managed hubs, I would go insane.  Since the
admins are the only ones ever dealing with these devices directly, I
think the argument is stronger for machine-generated names in their case.
And again, you can always add the alias for "bigbertha" or whatever.


Xev> And very importantly, document and publish both the scheme and why
Xev> you chose it.  If you don't do this, someone will come along 6
Xev> months later and try to do it all over again.

I hate to say it, but even if you DO document everything, someone will
STILL come along 6 months later and try to do it all over again.  :-)


Xev> [Note that my views do not reflect my employer, the people that I
Xev> work with, or anyone that I've ever met. They are solely based on
Xev> years of bashing my head against others on this issue].

Same here.  Happy bashing.


=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=
Paul Caskey            pcaskey@swcp.com            http://Paul.Caskey.com
=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=
  "People demand freedom of speech as a compensation for the freedom of
             thought which they seldom use."  --Kierkegaard