Sean Durkin:
[ Charset windows-1252 converted... ]
> Hi Viktor,
>
> Am 10.09.2014 um 16:19 schrieb Viktor Dukhovni:
> > Have you tried disabling TCP window scaling? It might be confusing
> > some middle-box (firewall, NAT device, ...) on path between the
> > remote systems and your MTA.
> I would not have thought of that... I've tried that now, but it does not seem
> to help.
>
> > Post the hostname/IP address of the receving system.
> mail.tuxroot.de
>
> > Capture and examine a tcpdump recording of mail from one of the
> > problem senders. Any sign of retransmission by the sender?
> I'm trying to get a good dump and will post results once I get one.
> Not that easy since the external hosts keep changing all the time. All mail
> affected comes from mass mailers that use server clusters, so I keep getting
> those messages from lots of different remote hosts. I'm waiting for it to
> happen from one of the hosts I've seen before.
>
> Retransmission is tried numerous times, but for every retransmission
> the lost connection message is the same (identical number of bytes),
> as far as I can tell. That's one thing that puzzles me... So e.g.
> a message is delivered twice and each time the connection is lost
> after exactly 17441 bytes, even if it's different remote hosts
> trying, that's kind of odd.
No, it means the same problem is happening. Same error,
same symptom.
> What I do notice there is that in fact the connection seems to be
> *very* slow. In the above example, the whole process takes several
> minutes. I don't have any throughput or network speed issues with
> other hosts, though. I've tried sending mail from Gmail, Yahoo,
Slow performance is typical for TCP window scaling problems. Have
you tried to turn it off in your kernel?
# sysctl -w net.ipv4.tcp_window_scaling=0
To make it permanent:
# echo 'net.ipv4.tcp_window_scaling = 0' >> /etc/sysctl.conf
Wietse
