> 1) You don't have to screw with any of the gateway software already in place.
> The WISCVM stuff is screwey enough *without* having to worry about
> repackaging some files, maybe, if it's needed...
That's right Valdis. Agreed on this one.
> 2) You know beforehand who is on the list. All I'd have to do is get LISTSERV
> to not run lists flagged as being in a certain class (junk mail, off-node,
> whatever). (Hint - this is a mod request... :-)
> If all else fails, I can always just run those lists off a second LISTSERV
> which would get AUTOLOGGED late at night.
Or you could have VMUTIL issue HOLD and FREE commands for the lists at schedu-
led hours. This is not much of a problem.
> 3) You don't have to worry about the digest at the other end making up the
> distribute list. I know for a fact that this could get VERY hairy at the ARPA
> end, as currently the digesters mail one copy to each person on the list. It
> would require that for DISTRIBUTE, the digester program flag all BitNet users
> and lump them for special processing. With a mailing list, they just have to
> add the one address to their target table, and remove all the .BITNET user
> addresses. No hassle.
Sorry Val. In my original idea the gateway creates the DISTRIBUTE job. That
means the ARPA people have nothing to do at all -- the BITNET people get sent
to the gateway in a single BSMTP envelope as usual.
> 4) LISTSERV will make a user subscribe to the appropriate server. Just have
> the digest lists at a fair number of places, with open subscription.
This one is irrelevant. DISTRIBUTE would route their copy of the mailfile to
the appropriate server, automatically.
> 5) With a mailing list, control is back in the hands of the postmaster, as
> some sites have requested.
Yes, *but* you have only one-tenth of the servers participating since creating
distribution lists for ARPA digest is a real pain considering the number of
lists. It also takes up space on the LISTSERV disk, by the way. About 5 sites
currently redistribute ARPA lists on a total of 41 servers. That would not be
a major reduction...
Some people, a lot of people, might be willing to take on ARPA redist but
might not have the time to create 30 lists. I'm thinking about the central
nodes whose CPU is (partly) dedicated to the network but who have a very limi-
ted staff.
> 6) With a mailing list, we can fix it so that MAIL can read it, and probably
> also fix the Reply-To: to point back at the master list on the ARPA side so
> that their reply gets put into the digest rather than floating around Bitnet.
> This is currently a flake with the Unix-Wizards digest, where a lot of the
> postings arrive seperately, because they did not go through the digester.
Yeah, *but*: I keep getting complaints from at least five people that my LIST-
SERV munges the nice, sweet "Received:" headers that are soooooooooo important
(those people are still a minority of course). Since I don't see any real use
for the "Received:" tag FOR BITNET ORIGINS, except to waste space on your disk,
I don't listen to these people. However, in the case of an ARPA address it
might help locating the guy. Don't flame me on this, Valdis, better send the
flame to these people directly ;-)
Another technical point about DISTRIBUTE: let's assume you are node xEARN,
the central node of country X. You're receiving a 1000-lines DISTRIBUTE on your
LISTSERV for, say, four people of your country, and you're concerned with the
amount of resources it will eat. Basically LISTSERV eats:
- Spool space
- CPU time
- Network resources
With DISTRIBUTE four files are sent out (or less if there is another LISTSERV
down the path) and one is received. Without it, the files go through your node
anyway, four are sent out, four are received. It obviously eats more network
resources.
The spool use is the same since in both cases, four files are sent out.
LISTSERV would eat about 15 seconds CPU to process the DISTRIBUTE job. 15
wasted seconds. However your RSCS will have received only 1 file instead of 4.
That's about 5 seconds CPU saved. So on the average you'll have paid 10 secs
CPU to save yourself 3000 lines input. Note that other people will have wasted
some of their CPU for you too. It's like RSCS, it works both ways. You don't
have any control over DISTRIBUTE, but you don't have any more control on RSCS.
If RSCS was intelligent enough to automatically 'fork' files, DISTRIBUTE would
not even be required! :-)
Eric
|