FYI -- my response (see original below) / pmw
 
  - - The original note follows - -
 
Date: Thu, 20 Jun 1996 10:45:48 EDT
From: Peter M. Weiss <[log in to unmask]>
Subject: Re: Studies on Internet Reliability
 
I too run a list that has approx 5k users on it.  Distributions
are about once per three weeks.  I would say that about
one per cent bounce even after I've cleaned up the carnage
i.e., list maintenance, from the previous distribution.
 
Lot's of unknown userids (real unknowns or temporary because
the target's authentication system db was down), unknown
hosts (temporary or permanent because one of the authoritative
DNS was fubar), host config points to itself (didn't they
notice that no one was getting e-mail????).
 
--  co-owner:   INFOSYS, TQM-L, CPARK-L, ERAPPA-L, JANITORS, LDBASE-L, et -L
URL:mailto:[log in to unmask]       "Do you suffer from analysis paralysis?"
 
<original posting ...>
 
Date: Wed, 19 Jun 96 09:53:00 CST
From: [log in to unmask]
Subject: Studies on Internet Reliability
Sender: [log in to unmask]
 
I am presently taking a graduate course towards my masters in
telecommunications management'.  I am looking for any studies that
have been done regarding the reliability of Internet email or other
measures of Internet reliability.  One of the types of data might be
if a thousand messages were sent, what per cent arrived within one
minute, within ten minutes, within an hour, within a day, were not
received, etc.  Any other measures of reliability would be helpful as
well.
 
If anyone knows of any data like this, I'd be grateful if they could
email me at: [log in to unmask]
 
Thanks in advance for your assistance.
 
 
Fred
 
 
[TELECOM Digest Editor's Note: I can tell you from my own experience
with the Digest mailing list which has about five thousand names at
any given time I get back 150-200 mail daemons from each issue (if
I do five issues that day that means about a thousand daemons) which
indicate trouble was encountered in delivery. Deducting the 'user
unknown' daemons which are the only ones I can do anything about (I
remove those names from the list after two or three such returns on
any given name, and there are usually two or three such removals
daily), I still get a good number of daemons which complain about
delivery problems. Typically the problem will be 'host unknown' or
'host not receiving mail, will keep trying' or 'cannot resolve host
name'. Sometimes the name resolvers are hosed; that is they got a
bit scrambled up and for the time being don't know where to go to
find (or seek resolution for) a given domain or site, etc.
 
A dozen or so daemons each time say 'timed out during user open with
site.name' and this generally means when the delivering site went
calling on the receiving site it got there and 'knocked on the window'
(connected to the appropriate socket) and the receiving site 'opened
the window' alright, but got impatient and slammed the window shut
before delivery was completed. Or so the sending site thought ... or
maybe the delivering site stood there for the longest time waiting for
the receiving site to accept the mail and finally gave up and went
away leaving the recieving site still waiting for more mail, etc.
Maybe the item of mail did not terminate to the satisfaction of the
receiving site, etc. In all probability the mail did get delivered
just as it gets delivered eventually to places where initially the
report was 'host unknown'.
 
Typically an issue of this Digest takes about 18 hours for complete
delivery to all names on the list, and I have the list sorted from
the '@' sign to the right so that all users at a given site get
serviced at the same time, which sendmail would do anyway but by
sorting it in advance sendmail does not have to 'work so hard' to
find everything going to the same place; that sorting seems to cut a
couple hours off the total time. At any given time around the clock
usually I have four or five instantiations of sendmail running; if
I start them sooner than about an hour apart (like say two within
thirty minutes of each other) they will 'race each other' as they
each benefit from the hostname lookups the other one is doing and
this causes some people to get a later issue of the Digest a few
minutes before they get an earlier issue (i.e. one does a lookup
and has it in the place where those things are stored; another
sendmail comes along looking and finds the information the first one
had squirreled away and uses it also). Typically the mailqueue
has ten to twelve thousand names waiting for delivery when I am
working on several issues in a day's time. I just now did a
check on it with the command 'mailq | wc' (meaning just count
the lines, don't send me the whole %$$#* stdout) and the response
was 8708 deliveries pending. Sometimes I forget the '| wc' part
of that request then I sit here and cuss while screen after
screen is dumped at me. <grin>  I encourage everyone to learn about
sendmail and how it works. The documentation is only about two
thousand pages long or the size of an Ayn Rand novel and good
for reading before bedtime.  <more grin>     PAT]