Multi-WINS behavior.

[Peter Samuelson]

[crh]
> I also discovered that I didn't have to jump through all of the
> foolish hoops just to get the C compiler licensed.  Once I was able
> to repair the bad disk I copied the licenses and they worked.

Yes, flexlm seems to be rather sane as license managers go.  By
contrast, although netls/iforls/lum/whatever-it's-called-this-week
works ok for node-locked licenses, their floating license server sucks
pus through a capillary tube.  I once spent four hours on the phone
with IBM trying to get our license daemon back up and working.

Getting a bit off topic, though, aren't we?  (:

> > Yes, I know, different machines might see different parts of the
> > namespace, but at the heart of it all there is still only the One
> > True, Flat Namespace.
> 
> I see it as multiple, disjoint namespaces, but perhaps this is just a
> question of terminology.  I'm calling the broadcast domain a
> namespace, and the set of all names in the WINS server database is a
> namespace.  The namespace of a given node is the union of all
> namespaces in which it participates.

Yes, it's a matter of terminology.  To me, a namespace represents "the
range of machines across which a name collision would cause problems".
I still maintain that this means any machine reachable through NetBIOS,
be it via WINS, broadcast or lmhosts.

Now, not everyone will necessarily see the whole namespace at a time.
If you have WINS to bridge multiple broadcast domains, and not everyone
uses it, some machines will only see part of the namespace.  To me that
does *not* imply that what they see is a separate, smaller namespace.

Now my fundamental point: for any set of machines communicating through
NetBIOS, there is only one namespace (by my definition, above).  *Any*
name collision risks confusion (at the least), because there is in
general no way to specify that you want only a machine in a particular
sub-namespace (from a particular WINS server, or in the local broadcast
domain, or whatever).

Yes, you can run an H- or M-node and prioritize which sub-namespace you
want to see first, but if either the WINS server or the target machine
goes down, you'll still get the wrong machine.  This, to me, is a much
worse eventuality than getting a false negative lookup.

> > Maybe in another five years everyone will have migrated to Win2k
> > with DNS and we can all forget about this.
> 
> Trade the old problems for the new.

(:  The trade press loves Win2k but what would you expect.  My
experience to date:

One guy here wanted us to load Win2k on his laptop, for DVD.  Two days
later he brought it back, said to wipe it and put Win98 on -- it seems
he couldn't get any work done because Win2k kept crashing.  So we did,
of course, but meanwhile I'm sitting here thinking: let me get this
straight, you want Win98 because the alternative crashes too much? (:


> I'd just like to know how it's done.  MS uses port 42 for WINS
> replication.  This port is 'officially' assigned to an old
> nameservice of some kind.

It was "IEN 116", whatever that specified.  (I don't feel like looking
it up.)  But TCP or UDP?  Hmmm.  AIX /etc/services says UDP.  Linux,
Digital and IRIX all say TCP.  Windows NT says both.  HP-UX says
neither.  I guess it must be obsolete, as nobody seems to agree on
which protocol it is.

> I'm not sure what it was for, but MS apparently figured that it
> wasn't being used any more so they grabbed it.

Microsoft?  Picking and choosing which standards are worth conforming
to and which ones to trample at will?  Say it ain't so....

> Okay, how do you define "a given network"?  I do consider the
> broadcast namespace and the WINS server namespace to be disjoint,
> which would put M and H nodes into the 'crude hack' category (which
> might very well be the case, but...).

As above, "a given network" comprises everyone who can talk via
NetBIOS.  (And I don't count use of IP numbers or DNS.  Those are other
namespaces.)

Peter