RE: A philosophical question about large directory structures

"James Benedict" <[email protected]> Fri, 23 Jul 1999 11:24:51 -0400
Newsgroups gmane.ietf.lsd
Message-ID <[email protected]>
> -----Original Message-----
> From: Andrew Probert [mailto:[email protected]]
> Sent: Friday, July 23, 1999 11:09 AM
> To: Benedict, James [CAR:5N41:EXCH]
> Cc: Erik Skovgaard; [email protected]; Michel, Ian [FITZ2:8M52:EXCH]
> Subject: Re: A philosophical question about large directory structures
> 
> 
> I think you are omitting scenarios for larger companies like:
> 
>     [email protected]
> 
> Large corporations need to subdivide their name space so as to effect
> routing of messages within their own networks, which are
> geography-based.
> 
> If they happen to have Joe as a unique entry in an internally
> addressable directory they may allow [email protected] and resolve 
> the routing
> by an internal lookup.
> 
> This example is real.

Your example is real for Nortel as well, although we still generate
external email addresses that are [email protected] (allows us
to move email servers internally without having to reprint business cards :)

What I am trying to do is present a model that provides the easiest
intuitive
route to guess a DN.  If a company decides to subdivide their namespace to
effect geographic
routing then that is fine as well. Whether they provide directory services
at dc=us or not, they
will still have the dc=hp, dc=com.  

If we take the top-down model of directories, I get

dc=us, dc=hp, dc=com
dc=ca, dc=hp, dc=com
dc=it, dc=us, dc=nortelnetworks, dc=com
dc=it, dc=ca, dc=nortelnetworks, dc=com 
dc=hr, dc=ca, dc=nortelnetworks, dc=com
dc=hr, dc=us, dc=nortelnetworks, dc=com

which presents a tree like:

com
  +--hp
  |    +--us
  |    |    +--joe
  |    +--ca
  |
  +-nortelnetworks
       +--us
       |    +--it
       |    +--hr
       +--ca
            +--it
            |  +--grunt
            +--hr
  

I think this tree is more intuitive than:

o=hp, c=us
o=hp, c=ca
ou=hr,o=nortelnetworks, c=us
ou=hr,o=nortelnetworks, c=ca
ou=it, o=nortelnetworks, c=us
ou=it, o=nortelnetworks, c=ca

ca
 +--hp
 +--nortelnetworks
 |   +--it
 |       +--grunt
 |   +--hr
us
 +--hp
 |   +--joe
 +--nortelnetworks
     +--hr
     +--it

As to whether I can discover LDAP services at ldap.us.hp.com or at
ldap.hp.com or both is also
up to the company.


> 
> 
> James Benedict wrote:
> > 
> > Erik,
> > 
> > You are right about the size of dc=com, but I think its 
> size is caused
> > as much by the fact that we tried to divide DNS up by 
> country as well.
> > 
> > When you talk in internet terms, I don't think that 
> countries make much
> > sense.  This has always been a problem for large 
> enterprises like Nortel
> > Networks.
> > But, today, even smaller companies (less than 60 employees) 
> have a global
> > presence.
> > In a location-oriented scheme, where do we put companies like this?
> > 
> > The internet itself transcends geographical boundries, I 
> think its naming
> > contexts need to as well.
> > 
> > I'de like to re-visit the dc-naming model from an end-user 
> perspective.
> > 
> > DNS, LDAP & email Alignment
> > 
> >   The idea here is that my DNS entry maps to the RHS of my 
> email address,
> > and that they
> >   both map to the dc components of my DIT.
> > 
> >   (eg)
> >   rfc822 = [email protected]
> >   DNS = nortelnetworks.com
> >   dc=nortelnetworks, dc=com
> > 
> >   What can I do with this:
> > 
> >   If I know a user's email address, I can guess that his 
> PKI cert will be
> > somewhere in the
> >   subtree of dc=nortelnetworks, dc=com.  I could go out on 
> a limb and either
> > try
> >   [email protected], dc=nortelnetworks, dc=com
> >   or
> >   DN=userid=grunt, dc=nortelnetworks, dc=com
> > 
> >   Failing that, I could try to do some sort of search 
> filter with email or
> > userid.
> > 
> >   I could always take a chance that ldap.nortelnetworks.com 
> is a valid
> > LDAPv3 server
> >   and ask for its schema attribute.
> > 
> >   Chances are, if I know anything about an organization 
> (besides its name)
> > it will be either
> >   an email address (with a DNS name) or a web-page (which 
> also has a DNS
> > name).  Anything else,
> >   I will have to do a search which will likely result in one of the
> > previously mentioned pieces
> >   of information.
> > 
> >   Overall, I think that modeling DNS does quickly reduce 
> the search set in
> > most cases.  I can
> >   also see cases where browsing a country-oriented tree 
> might confuse more
> > than help: Which country
> >   do you think Nortel is in?  What if I want to lookup 
> someone at my local
> > IBM office?
> > 
> >   Personally, I like the "Keep It Simple" principle.  DNS could have
> > benifited from a little better
> >   organization (didn't they try country as well), but "If 
> you can't beat
> > 'em... join 'em"
> > 
> > James A Benedict
> > Advisor, IP Directory Systems Architecture
> > Carrier Packet Solutions
> > NORTEL NETWORKS
> > Ph:  (613) 763-3909
> > 
> > > -----Original Message-----
> > > From: Erik Skovgaard [mailto:[email protected]]
> > > Sent: Thursday, July 22, 1999 12:31 PM
> > > To: alexis.bor; 'Tim Howes'
> > > Cc: 'Thomas Lenggenhager'; ietf-lsd
> > > Subject: RE: A philosophical question about large 
> directory structures
> > >
> > >
> > > Alexis,
> > >
> > > I still feel very strongly that the dc= naming is a bad
> > > thing.  Look at the
> > > number of entries immediately under dc=com!
> > >
> > > The correct thing to do is to reduce the search set quickly
> > > (CS 101 as I
> > > recall) and the best way we currently have is to use the 
> division by
> > > country.  In practice, that is what is done many places
> > > around the world.
> > >
> > > Certainly, we can agree that not all directories will be
> > > interconnected,
> > > but to make the assumption that none of them will, is 
> plainly a bad
> > > assumption.  I think PKI will drive a lot of the 
> interconnection or at
> > > least coordination.
> > >
> > > Cheers,                 ....Erik.
> > >
> > > ------------------------------------
> > > Erik Skovgaard
> > > GeoTrain Corp.
> > > Enterprise Directory Planning Services
> > > http://www.geotrain.com
> > >
> > > At 16:37 99/06/28 -0400, Alexis Bor wrote:
> > > >Tim,
> > > >I agree with you that domain names are what people are good
> > > at guessing
> > > >these days.  I think that we learned that lesson at least
> > > twice, once with
> > > >email and once with the web.
> > > >
> > > >I have been preaching the idea of the DN as just a pointer
> > > to an entry since
> > > >the early 90s and have always hid them from all of my
> > > applications.  I'm
> > > >glad that others are seeing the value in that, but it has
> > > been a long road.
> > > >
> > > >Naming has always been an ugly thing and it has caused lots
> > > of grief to all
> > > >of us in the directory space, as well as others.  Everyone
> > > seems to take
> > > >names very seriously - and understandably so.  The real
> > > question is how to
> > > >do we get past that.  I think that dc naming gets us part of
> > > the way there.
> > > >What we need is a mechanism where if we don't know the dc,
> > > to go out and
> > > >find it - and hopefully we can do that with protocol most of
> > > the time.
> > > >
> > > >You made an interesting comment early on in this thread that
> > > 4 years ago you
> > > >realized that directories wouldn't be all connected and that
> > > it was probably
> > > >a bad thing if they were.  I tend to disagree with that.  I
> > > think that you
> > > >were right 4 years ago that the current method of doing it
> > > was not workable
> > > >for both social and technical reasons.  However, that
> > > doesn't remove the
> > > >need for doing it some day.  If you think of DNS as a form
> > > of directory,
> > > >then it is possible and is useful.  I think that there are
> > > many applications
> > > >that are currently disabled because they can't get to
> > > information that could
> > > >be stored in directories.
> > > >
> > > >We are on the verge of having directories deployed in almost
> > > every company
> > > >with an IT department on the planet, many of them will 
> have multiple
> > > >directories from multiple vendors.  At that point, it will
> > > be clear that we
> > > >need to quickly figure out how to do it, and the rewards
> > > will be great.
> > > >Just think back of what it was like to maintain HOSTS tables
> > > for every
> > > >system running IP, and that wasn't very long ago.  I think
> > > that twenty years
> > > >from now, we'll be talking about the good 'ol days when
> > > business to business
> > > >communication that uses directories was just a dream that
> > > many of us were
> > > >saying would never happen.  And noone will believe that they
> > > made routers
> > > >and hubs that didn't use DEN...
> > > >
> > > >-- Alexis
> > > >
> > > >
> > > >Alexis Bor
> > > >Directory Works, Inc.
> > > >P.O. Box 470276
> > > >Celebration, FL  34747-0276
> > > >407-566-9250
> > > >407-566-9251 (FAX)
> > > >[email protected]
> > > >http://www.directoryworks.com
> > > >
> > > >
> > > >> -----Original Message-----
> > > >> From: [email protected]
> > > >> [mailto:[email protected]]On Behalf Of
> > > >> Tim Howes
> > > >> Sent: Wednesday, June 23, 1999 6:58 PM
> > > >> To: Erik Skovgaard
> > > >> Cc: Thomas Lenggenhager; [email protected]
> > > >> Subject: Re: A philosophical question about large
> > > directory structures
> > > >>
> > > >>
> > > >> Just an opinion, but I don't think anybody wants to
> > > >> guess at distinguished names. I actually think it's
> > > >> likely to be significantly less effective than trying
> > > >> to guess domain names, which people seem to be guessing
> > > >> or otherwise figuring out today without too much
> > > >> trouble. And I don't think any respectable user agents
> > > >> will support guessing except possibly as some last
> > > >> resort mode of operation. We've seen directory clients
> > > >> moving more and more toward hiding DNs from users, and
> > > >> I think that's a very good thing and one we're likely
> > > >> to see increasingly.                 -- Tim
> > > >>
> > > >> Erik Skovgaard wrote:
> > > >> >
> > > >> > Thomas,
> > > >> >
> > > >> > I agree that namespaces have to be unique, but I *also*
> > > >> believe they should
> > > >> > make sense to users.  Granted, *some* host names are
> > > >> user-friendly, but
> > > >> > many are not.
> > > >> >
> > > >> > In the specific example you mention, I would suggest 
> that {o=Sun
> > > >> > Microsystems, c=ch} is a reasonable and guessable
> > > >> namespace.  I would not
> > > >> > venture to guess what your hostname is.
> > > >> >
> > > >> > Cheers,                  ....Erik.
> > > >> >
> > > >> > -----------------------------------
> > > >> > Erik Skovgaard
> > > >> > GeoTrain Corp.
> > > >> > Enterprise Directory Engineering
> > > >> > http://www.geotrain.com
> > > >> >
> > > >> > At 23:41 99/06/23 +0200, Thomas Lenggenhager wrote:
> > > >> > >On Wednesday, 23 Jun 1999, Erik Skovgaard writes:
> > > >> > >> The DC namespace is poorly suited for the Directory
> > > >> since the Internet host
> > > >> > >> names are not all guessable.  For instance, if I were to
> > > >> look up Canadian
> > > >> > >> Airlines, I would have to *know* that the hostname is
> > > >> 'cdnair.com'.
> > > >> > >
> > > >> > >We are going a bit off track with these messages about
> > > namespace.
> > > >> > >
> > > >> > >The namespace should mainly be unique. That would be
> > > >> provided by DC.
> > > >> > >Finding something should not be tried to achieve by
> > > >> guessing the right
> > > >> > >name, but by searching for it (index/directory).
> > > >> > >
> > > >> > >> Since the Directory name registration is closely tied to
> > > >> the legal names of
> > > >> > >> companies in most countries, I think that makes for a
> > > >> much better choice.
> > > >> > >> I would rather find Canadian Airlines as {o=Canadian
> > > >> Airlines, c=ca} than
> > > >> > >> the somewhat cryptic {dc=cdnair, dc=com}.
> > > >> > >
> > > >> > >And how about the official name of Canadian Airlines in
> > > >> French and the
> > > >> > >DN you proposed?
> > > >> > >
> > > >> > >What is the official name of a company?
> > > >> > >       Acme    or Acme Inc.    or Acme, Inc.
> > > >> > >
> > > >> > >Just as an example of quite ugly and non-guessable names,
> > > >> the official
> > > >> > >name of Sun in Switzerland:
> > > >> > >       Sun Microsystems (Schweiz) AG
> > > >> > >
> > > >> > >Thomas
> > > >> > >
> > > >> > >
> > > >>
> > > >> --
> > > >> Tim Howes
> > > >> Vice President, Technology
> > > >> Office of the CTO
> > > >> America Online, Inc.
> > > >>
> > > >
> > > >
> > > >
> > > >Attachment Converted: "d:\Program Files\Eudora\Attach\Alexis Bor
> > > (E-mail)3.vcf"
> > > >
> > >
>