Migration from Samba-TNG-Alpha 2.6.1 "back" to Samba-2.2 Version possible

[Luke Kenneth Casson Leighton]

On Wed, Oct 24, 2001 at 12:10:26AM +0200, Elrond wrote:
> On Tue, Oct 23, 2001 at 09:26:36AM +0200, Luke Kenneth Casson Leighton wrote:
> [...]
> > > I'm interested in a flat-file smbpasswd that actually does scale,
> > > which I still think is possible, with indexing.  Then again, I've never
> > > profiled it..
> > 
> > well the at least O(N^^3)algorithm on _top_ of the smbpasswd
> > db is far more serious a problem, which may be compounded
> > by the underlying db being O(N) or grearter.
> 
> One part of the n^3 (I believe, it's not as worth, but it's
> not really good) is due to the missing rid in smbpasswd.
 
that is absolutely correct.

the lookup procedure going from uid or gid to RIDs involves
going via the unix passwd and group databases using the
bloody _names_.

so not only do you get a hit for NT-name to unix-name
conversion, with all the problems and confusion 
associated with "domain user map" and "domain group map"
and "domain alias map" having absolutely nothing to do
with "user map" and [the unused] "group map",

but also you get at least an O(N**2) hit for doing from
uid or gid name to RID, oh, i can't quite decide
whether the name is now a unix group name or a unix
user name, oh, well, _let's_ look it up in _the entire_
SAM database to find out, shall we?

it's just a total mess, solved instantly and easily by
not associating with the unix database _at all_.

and the job of relating SAM RIDs/SIDs to unix uids/gids
is solved by SURS.

[no, jeremy, it's not solved by winbind.  winbind just happens
to provide one implementation of SURs, for people who
use nsswitch.]

luke