Re: Unsubscription bounces - sites sending unsubsription commands instead of ...

[Eric Thomas <ERIC @ SEARN . SUNET . SE>]

On Fri, 28 Apr 1995 23:51:51 -0400 PMDAtropos@aol.com said:

>In fact, if we  take America Online as an example, it  would be great to
>be= able to  have the service mail system automatically  issue SIGNOFF *
>(GLOBA= L to LISTSERV for extinct user accounts.

With the  amount of users you  have this would kill  LISTSERV immediately
and  I would  have to  distribute an  emergency patch  to eliminate  this
command once and  for all. I've come  this close to doing  it for several
releases. Eventually I will, it's only a matter of time. This command was
developed a  long, long time ago,  when the network was  much smaller. It
doesn't  scale up  and it  cannot be  made to  scale up.  You can't  have
thousands of  sites (approximate number of  universities) generating tens
of thousands of netwide signoff  requests each a year (approximate number
of accounts that expire)  to tell the world that, in  case for any reason
they had Joe  on their list, well,  Joe is gone for good.  That's tens of
millions of Joes a  year for each server to look up or  just 27k per day.
If you're willing to sacrifice 10%  of your entire machine to that, which
I  doubt most  people would,  the queries  have to  complete in  0.3 sec,
counting everything  - SMTP  receipt and  all. But  in reality  that's an
optimistic  scenario.   All  the   academic  account  removals   will  be
concentrated at  semester end. So in  fact it's more like  you'll get the
10M Joes in the space of one month or  > 300k a day and 30ms each. If you
have a program that can run SMTP at that rate, I have a job for you :-)

The only reason netwide SIGNOFF isn't doing any of that is that virtually
noone uses  it because virtually  noone knows about  it. It remains  as a
kind of "privilege"  for the LISTSERV users who were  there in the BITNET
days and know  about it. Because, indeed, it is  convenient. The only way
to make it scale up would be  to have each organization run a server that
stored  the subscriptions  to all  its users  all over  the world.  These
servers would run  the search and extract the 0.1%  of matches and notify
just the  servers in  question, indicating which  lists are  affected. It
still wouldn't work for concealed subscriptions, though.

>2,000 AOL-related bounce messages a week if I don't keep right on top of
>things  -- that  2,000/wk figure  is about  20% of  the total  amount of
>bounce=

The  real problem  is  that AOL's  bounces  are not  parsed  by the  mail
managers. If they were, the users would get deleted and everyone would be
happy. And this scales up because only users actually on the lists result
in a delivery error.

  Eric