Note: I'm not on XMAILER nor MAIL-L (too much noise). If you want me to get
your replies, make sure to include me as a primary recipient.
>> Negative comments from someone on Internet:
>>
>> > > After the (latest) flood of bounced mail from BITNET, I really have to
>> > > wonder if there is any point in setting up regional BITNET gateways.
>
>What is the latest flood? The last problem I've heard of was in the BITNET
>redistribution of INFO-HAMS -- the problem there was PMDF (which I believe is
>more common on the Internet than on BITNET!) sending the rejection to the
>Reply-to instead of the Sender. From what I heard, the problem wasn't even a
>mail loop because LISTSERV successfully intercepted the (re)distributed
>rejections.
So it would seem the problem was not a problem at all?
>> > > BITNET nodes seem to be a world unto themselves, and tend to be much
>> > > less reliable than nodes on just about any other "network" around
>> > > (including USENET).
>
>It is easy to cast aspersions on the many for the sins of the few. I don't
>think it's useful, however.
That's a funny remark. I've been on BITNET for over 2 years, and receive some
75-100 files a day. I've only lost files twice in this period (not counting
files sent to nodes which had just connected and didn't have their software
properly installed). However, less than 50% of the mail I sent to UUCP went
through. The rest simply got me rejection notices with "uuxqt" subjects and
"unknown imail error 1". I guess it all depends on the sunglasses you are
wearing when looking at other networks' reliability.
>
>> Attempted justification for unreasonable behavior on BITNET:
>>
>> > As I understand it, the reason that LISTSERV sets the Sender field to the
>> > name of the list instead of some human being's mailbox (as RFC822 says to
>> > do) is that the Rice Mail user agent (and maybe others too?) uses the
>> > Sender field to decide where to file messages.
>
>That is not quite true. What user agents do, by and large, is file messages
>according to the external tags (the "filename"). The real problem is that
>LISTSERV uses the Crosswell MAILER for sending files, and MAILER makes the
>filename match the Sender field. Also, it turns out that a human couldn't
>tolerate being listed as the Sender for any reasonably busy LISTSERV because
>of the incredible flood of messages back to the Sender reporting on progress
>of the mailing.
A file called "LSTSRV-L MAIL" with a "Sender:" of LSTSRV-L@FRECP11 is mail
from the LSTSRV-L list. A file called "ERIC MAIL" with a "Sender:" of
ERIC@FRECP11 would be mail from... maybe me, or maybe LSTSRV-L, or maybe
RSCSMODS, or (...) or anything else. To know what it is about, I have to LOOK
into it.
>> > LISTSERV has filters that most of the time are able to tell that a
>> > message sent to a list is bounced mail and should not be distributed, but
>> > they aren't 100 percent successful, so some do get distributed to the
>> > lists.
>>
>> Now, for my comments: (I am generally biased in favor of BITNET, if the
>> poor connectivity problem can be remedied)
>>
>> (1) The fact that Rice Mail uses the Sender to determine where to file the
>> mail is irrelevant. Fix Rice Mail (which apparently has a prereq of a
>> change to LISTSERV). The fact that it has a problem *does not* justify
>> doing something IN VIOLATION of the STANDARD (RFC 822) that is
>> supposed to govern all the mail.
Cool down guy. The mail items generated by LISTSERV contain perfectly valid
RFC822 headers, ok? Just because YOU think that LISTSERV is a mailing agent
does not mean that *I* think so. I see LISTSERV as a PROGRAM, which receives
an implicit command saying "please distribute this piece of mail to the
following persons, creating new headers according to the format they have
specified in their distribution options". The default format is to trash all
the 'Received:' junk and suchlike, which most people are very happy with. You
can also ask to keep all the aforesaid junk if you want. The list owner may
specify that a given address be put in the "Sender:" field, thus he can create
mail with a "Sender:" pointing to himself if he suffers from masochism. If he
doesn't, it's not my fault.
>As noted above, the problem is more one of MAILER than of any user agent. I
>have passed a detailed suggestion to Alan Crosswell for resolving the
>problem. It should be noted, of course, that MAILER
Good luck. I personally won't be angry with Alan if he ignores your
suggestion.
>isn't in any sense "at fault", just not sufficiently flexible for the needs
>of LISTSERV. For that reason, it is unfair to point a finger of blame in
>*any* direction, since MAILER and LISTSERV are maintained independently.
Right.
>> (2) Those filters really bother me. The fraction of a percent failure rate
>> has wreaked havoc on the network multiple times in the last year. The
>> filters are a band-aid solution to a problem caused by other pieces of
>> software not being reasonable. Fix the other software and remove the
>> filters.
Sure. Start fixing the 500 VAX/VMS mailers which send rejection notices to the
"Reply-To:" field first, and we'll continue the discussion when this has been
done.
>Wait! Don't remove the filters, whatever else you do! There will always be
>unreasonable software somewhere in the world.
Especially gateways. Look at the BGUNOS one - the worst I've ever came afoul
of in BITNET. Oh well.
Several people asked me why LISTSERV doesn't integrate the functions of
MAILER. The answer is twofold:
1. A mailer must be written in assembler, or possibly in a compiled language
like FORTRAN (yuck!!!) or PASCAL (have fun on the I/O business) or C (sigh)
or PL/1 (be sure that all the potential users have enough $$$$ to buy
themselves the PL/1 library license - it's not PD). The reason why LISTSERV
is written in REXX is that it can work on any VM/SP site without any
additional license, and the resulting code is small.
2. A mailer must, for obvious technical reasons, run with a different "input
mailbox" than LISTSERV. Mail sent to LISTSERV is normally a set of commands
to be executed and returned to the sender. Mail sent to MAILER is something
to be distributed. LISTSERV cannot assume that stuff arriving in its
mailbox must be distributed unless it points to the LISTSERV userid,
because shitty UUCP gateways often place mail in LISTSERV's reader saying:
"To: xxyw!zyueez!uiu!(...)!frecp11!listserv" (something with an '@zzzz'
appended to it, but not always). Gateways often don't bother to translate
the 'To:' field. I would be awfully glad to trash this out as being invalid
RFC822 mail, but then users would complain and UUCP gateway administrators
are quite a bit anti-IBM and refuse to "make changes for shitty
brain-damaging IBM machines".
Anyway a MAILER must have a very good response time, which is not possible
inside LISTSERV since other time-consuming commands might be executing when
the mail request is received. Keep in mind that the EARN-JANET gateway,
running on an IBM 3081-K (lots of MIPS), is idle 0% of the time during peak
hours, and it's written in assembler+C. That is, enough files are coming in
for it to be busy 100% of the time. If it were written in an interpreted
language, it would NEVER be able to clear up the queues.
Eric
|