From: John Gilmore <gnu@toad.com>
Date: Tue, 11 Feb 1997 15:10:22 -0800 (PST)
To: Patrick May <pjm@spe.com>, gnu
Subject: Re: Moderation experiment almost over; "put up or shut up"
In-Reply-To: <199702112210.OAA01556@gulch.spe.com>
Message-ID: <199702112310.PAA28883@toad.com>
MIME-Version: 1.0
Content-Type: text/plain


>      - Is a pentium up to the task of running a list of this size and
>        volume?

A pentium is definitely up to this task.  I've been running it the
whole time on a slower 40MB SPARCstation-2 (that also runs netnews and
general computing).  Give it a big /var/spool partition (mine is 60MB)
because every message will sit in the queue for days (*somebody* on
the list will have an unreachable name server or MX server until the
msg times out).  Give it lots of RAM and paging space, since each
sendmail process takes about 2MB virtual, 1.4MB physical, and you will
have dozens running at the same time.

The new version of majordomo (that allows confirmation of
subscriptions) will help a lot.  It needs a small patch though, to do
exponential backoff on the lock file, or when you get a flood of
messages, thirty majordomo processes will burn up the whole machine
trying and failing to get the lock file.

You'll need a BIG mailbox for the bounce messages, and someone (or
some unwritten software) to scan it every day or two and delete the
lusers whose mailboxes are full or who dumped their account without
unsubscribing.  The bounce mailbox on toad gets between 1 and 4MB of
email a day; more when the list is under attack.

You'll want to run the latest version of BIND on the machine, too,
since doing DNS name-lookups on a thousand email addresses is expensive.
You want them all in the in-memory cache on the same machine.  The
name daemon burns about 7MB virtual, 5MB real RAM once its cache
gets loaded.

Make sure that every message sent to the list gets into at least
two logfiles -- on separate partitions, in case one fills up.  At
least if you want to have an archive of what's been sent.

>      I can provide a pentium box running Linux with a T1 connection to
> MAE-West to host the list, if there is still interest.

Make sure you are getting "transit" service to the Internet, instead
of trying to cheap out with "peering" to a few major networks.
Without transit service ("we'll carry your packets to anywhere even if
the destination is not on our network") you won't be able to route
packets to some places on the net.  This will cause mail to those
subscribers to sit in the queue for days and then bounce.

The real issue is how willing you are to put your own time into
dealing with problems.  Not only do things go wrong by themselves, but
there are malicious assholes in the world who will deliberately make
trouble for you just because they like to.  Spending a day or two
cleaning up the mess is just part of the job.  Check your level of
committment two or three times before taking on the task -- so you
won't end up getting disgusted after a month or two and putting the
list's existence into crisis again.  It's not a "set it up and forget
it" kind of operation.

	John