[comp.mail.sendmail] campus mail service

RThirlby@MULTICS.LOUGHBOROUGH.AC.UK (02/08/89)

Is it possible to include code in a sendmail system to fuzzy match
incoming mailnames using (say) the fields of the password file as basic
data.  we are in the middle of procuring a new main campus general
purpose and mail system, the only systems on the shortlist are all UNIX.
We have about 4000 registered users and expect this to increase with a
better mail service.  I am frightened as to the size and search time of
an alias file with several entries per user.  We clearly need to use the
alias system for users whos mailboxes are elsewhere but feel that an
algorithm approach should be possible for users with local mailboxes
given a sensible choice of username strategy.

We cant be the first to do this sort of thing???

Rob Thirlby, loughborough Univ Computer centre.

Andrew.Findlay@brunel.ac.uk (Andrew Findlay) (02/08/89)

| Is it possible to include code in a sendmail system to fuzzy match
| incoming mailnames using (say) the fields of the password file as basic

Yes. The IDA enhancement kit provides that service, along with a lot
of other nice things. At Brunel we use the IDA facilities to replace
usernames with full names in mail addresses.

| data.  we are in the middle of procuring a new main campus general
| purpose and mail system, the only systems on the shortlist are all UNIX.
| We have about 4000 registered users and expect this to increase with a
| better mail service.  I am frightened as to the size and search time of
| an alias file with several entries per user.  We clearly need to use the

All versions of Sendmail that I know of use a DBM (Hashed key)
database to store the aliases. Lookup is very fast.

| alias system for users whos mailboxes are elsewhere but feel that an
| algorithm approach should be possible for users with local mailboxes
| given a sensible choice of username strategy.

There are arguments both for and against fuzzy matching. It is
generally thought best to bounce the message if there are several
possible matches. Many people argue that ONLY exact matches should be
delivered, and all fuzzily-matched near-misses should be reported to
the sender with some helpful information to enable them to get the
address right next time.


| We cant be the first to do this sort of thing???

Not at all, but relatively few sites are putting much effort into
using the facilities that are now available.

I hope to have a version of Sendmail with IDA ready fitted available
for public NIFTP in the UK quite soon. This will be based on
Sendmail-5.61, which is the latest and greatest from Berkeley (and is
much more secure than the versions supplied by most manufacturers)


Andrew

---------------------------------------------------------------------
|  From Andrew Findlay at Brunel University, Uxbridge, UB8 3PH, UK  |
|  Andrew.Findlay@brunel.ac.uk          phone: +44 895 74000 x2512  |
---------------------------------------------------------------------

taylor@cs.glasgow.ac.uk (Jem Taylor) (02/08/89)

|>Is it possible to include code in a sendmail system to fuzzy match
incoming mailnames using (say) the fields of the password file as basic
data. <<

Sendmail is not the place to do this. This is a delivery agent task; once
the message has been selected for local delivery by sendmail, it is passed
to the delivery agent (usually /bin/mail); any program implementing
more intelligent delivery than /bin/mail is capable of should be declared as
the 'local' mailer and can than itself call /bin/mail for final delivery.

Having said that, some people have used IDA databases called from sendmail
to perform more complex transformations of user names than /usr/lib/aliases
allows...

Apart from that, it is generally considered unwise to deliver as the result
of a fuzzy match; instead you should return an error message listing possible
names which are 'near' the failed name, if an exact match is not found. What
constitutes an exact match is up to you of course ... but should be resistant
to changes in the user population. A name which resolves to user A should not,
next day, resolve to user B if user B is created over night.

-Jem.

cfe+@andrew.cmu.edu (Craig F. Everhart) (02/11/89)

We do fuzzy matching for 8000+ users in the Andrew Message System (well, in its
delivery system), too, which is at the moment available only as part of IBM's
AFS (Andrew File System) experimental deployment.  Sigh.

                Craig Everhart

Andrew.Hilborne@newcastle.ac.uk (Andrew Hilborne) (02/14/89)

In article <890208080455.315886@UK.AC.LOUGHBOROUGH.MULTICS>
	RThirlby@MULTICS.LOUGHBOROUGH.AC.UK writes:

> Is it possible to include code in a sendmail system to fuzzy
> match incoming mailnames using (say) the fields of the
> password file as basic data.
...
> Rob Thirlby, loughborough Univ Computer centre.

Incredibly sendmail already tries to do fuzzy matching against user names
in /etc/passwd.  There is a comment in the code to the effect that
sendmail already spends much of its time in the fuzzy-matching routine
which could do with being optimsied!

--
Andrew Hilborne, MARI Ltd, Newcastle upon Tyne, UK.