Quibbles re WINS

[Nicolas Williams]

The cache coherency problem could be fixed by ignoring NetBIOS
registrations and relying only on information from dhcpd (and assuming
single-homed clients; DHCP doesn't make it easy to distinguish client
moves from multi-homed clients...)...

:)

Or nmbd could use output from dhcpd to invalidate cached entries, rather
than ignoring NetBIOS registrations.

All of which is easier said than done.

Still, cacheing with *very* small TTLs should still be safe, no?

Nico


On Tue, Feb 09, 1999 at 06:31:40AM +1100, Luke Kenneth Casson Leighton wrote:
> On Mon, 8 Feb 1999, David Collier-Brown wrote:
> 
> > I asked
> > > >       Did you mean client or server?
> > 
> > Luke Kenneth Casson Leighton wrote
> > > > nmbd-as-a-wins-server.
> > > not a good idea, trust me.
> > 
> > 	Ok, I certainly believe you, but could you give the list a hint
> > 	about why?  It's highy counter-intuitive
> 
> if you have cache coherency problems due to registration of netbios names
> getting out of sync across a duplicated back-end (e.g ldap) it will be
> exacerbated by nmbd cacheing.
> 
> let's say that you have dhcp feeding the ldap database on one host, it
> duplicates to another.  you switch off and on again, you get a new ip from
> dhcp, it replicates to the other ldap server but you cached it.
> 
> let's say that you switched off rather than shut down, that means that a
> "netbios de-registration" will never occur, that means that the other ldap
> server never knows the name has been de-registered, nmbd never gets
> informed, it caches the old ip.
> 
> result: users of the other ldap server via nmbd get the wrong ip.
> 
> this is just one example of a potential horrible situation.  not having
> nmbd wins server cacheing eliminates a source of potential problems.
> 
> luke