Path: utzoo!mnetor!uunet!husc6!yale!cmcl2!brl-adm!adm!SRA@XX.LCS.MIT.EDU
From: SRA@XX.LCS.MIT.EDU (Rob Austein)
Newsgroups: comp.mail.headers
Subject: some mailers exhibit the following behavior
Message-ID: <11821@brl-adm.ARPA>
Date: 14 Feb 88 21:20:20 GMT
Sender: news@brl-adm.ARPA
Lines: 28

Well, while Jordan's algorithm probably doesn't exactly describe any
known mailer, it certainly describes a common scenario, where a mailer
(TOPS-20 or otherwise) picks the "best address" of the host it wants
to talk to and gives up on that host if the "best address" doesn't
answer.  This is obviously not entirely satisfactory, but the
alternative is generally accepted as being worse; trying multiple
addresses means that the mailer has wait for connection more than once
before deciding to give up on a host that is probably down.  For the
record, some of the algorithms currently used in ITS COMSAT for
delivering mail to hosts on the MIT Chaosnet require this kind of
multiple wait; I can assure anyone who doubted it in the first place
that this does horrible things to mailer performance.

The problem is primarily one of tuning the mailer to the application.
A mailer running on a personal workstation might well want to try all
the addresses of a destination before giving up, while a mailer daemon
running on a machine that is used primarily for relay/redistribution
(ie, massive amounts of mail neither originated nor destined for the
local machine) might well want to do what most mailers do now and give
up on a host after the first connection failure.  There is no single
right answer, it depends on the mailer's owner's priorities.

We (MIT-LCS) have found the distinction between endpoint and relay
mailers clarifies some of the issues involved in configuring a
workstation/server environment; perhaps mailsystem designers should
also give some thought to the distinction and its implications.

--Rob