Skip site navigation (1)Skip section navigation (2)
Date:      Sat, 23 May 1998 19:30:59 -0700 (PDT)
From:      "Jonathan M. Bresler" <jmb>
To:        andre@pipeline.ch (IBS / Andre Oppermann)
Cc:        grog@lemis.com, freebsd-chat@FreeBSD.ORG
Subject:   Re: FreeBSD mailing lists with high latency
Message-ID:  <199805240230.TAA14659@hub.freebsd.org>
In-Reply-To: <35670285.B6B73368@pipeline.ch> from IBS / Andre Oppermann at "May 23, 98 07:08:21 pm"

next in thread | previous in thread | raw e-mail | index | archive | help
	you all got a head start on me in this discussion.  ;)

IBS / Andre Oppermann wrote:
> Greg Lehey wrote:
> > 
> > On Sat, 23 May 1998 at  2:09:50 +0200, IBS / Andre Oppermann wrote:
> > > I'm now since Aug '96 on various freebsd-* mailing list and I see the
> > > latency between sending the mail and getting it back via the list
> > > is longer every day. In case of my last posting it took 1 hr and 20 min
> > > to get back.
> > >
> > > Is this related to the box the list is runnig on or is it simply
> > > an overloaded sendmail. Then it might be worth a thought to switch
> > > to qmail (it's really some hundred times faster than sendmail for
> > > big lists).
> > 
> > jmb will probably get back to you with more details, but briefly:
> > 
> > hub sends the messages to a number of different relays rather than
> > directly to each recipient.  Some of these relays can be slow.  To
> 
> I get my mails directly from hub.FREEBSD.org.

	that's correct, we dont have a mail-relay for .ch.
	any volunteers?  for other top level domains we have
	a number of relays....edu and com are excellent examples.

> 
> If I interpret the header correctly then this happens:
> 
> 1. sendmail on hub.freebsd.org receives the mail from the sender MTA
> 2. then it goes to majordomo on hub
> 3. majordomo expands to the list
> 4. majordomo sends the stuff to bulkmailer
> 5. bulkmailer sorts the recipients
> 6. bulkmailer injects all that stuff into sendmail for delivery
> 7. finally sendmail delivers the messages to the recipent MX host
> 
> > know where the delay is in your case, look at the headers.  For
> 
> The problem is the outgoing queue of sendmail (it rested there for
> 1hr and 15min).

	depends upon the day and the traffic that we experience
	as well as our connectivity to your site.

	recently we have had a number of bad days with lots of mail
	stalling on hub.....other sites being off the 'net unable 
	to receive the mail.  at times we had upward of 16,000 
	messages in the queues.  part of this is my fault for allowing
	people as much as 5 days before bouncing their mail.

> 
> > example, yesterday I got a couple of duplicate messages sent to two
> > different mailing lists.  It appears that the relay host had gone down
> > in between, and as a result the two copies appeared something like 10
> > hours apart.  The headers showed that they left hub a few seconds
> > apart.
> 
> I get it directly from hub.
> 
> > I don't know qmail, but I find it hard to believe that it is really
> > "several hundred times" faster.  It's possible that it looks that way
> 
> It is really much faster than sendmail. Sendmail has big problems with
> huge queues.
> 
> > due to different delivery strategies (try to deliver to all recipients
> > at the same time), but I would have my doubts about how effective this
> > is with 800 messages being relayed at the same time, as is currently
> > the case at hub.
> 
> Qmail, with concurrency limit set to 60, delivers those 800 messages
> in a little bit over 2 minutes (10 sec av. per connection) running
> on a Pentium-100 with 16MB. You can even put up concurreny up to 150
> with a little bit more RAM. I've even heard stories about delivering
> to over 60k remote addresses in 1.5 hours using stock qmail 1.01 on
> one P5 with BSD/OS with T1 connectivity.

	if i remember rob kolstad's paper correctly, sendmail will 
	process around 250,000 messages per hour, if the receiver 
	will accept the mail...that is a big if for us.
> 
> The qmail mailing list has something about 800 subscribers and I get
> back my mail via the list in under 80 seconds.
> 
> Sendmail has big problems if some MX for lots of messages in it's 
> queue are unreachable.

	sendmail's .hoststat cache addresses this issue.
	and may be the crux of your problem.  if we tried to reach you
	and failed, for any reason, we will not try again for at least
	30 minutes.  without message-id, i dont want to grub thru
	the logs looking for this.

	if we move away from sendmail we will move to vmailer from
	wietse venema, the author of tcp wrappers.  you can read
	about his mailer at http://wzv.win.tue.nl/vmailer/.

	its not that i have anything against qmail.  wietse is someone
	whose code i am more comfortable with, thats all.


	so why dont you send me some headers?
jmb

To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-chat" in the body of the message



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199805240230.TAA14659>