On Tue, May 07, 2013 at 10:59:45PM -0700, mailtime wrote:
> It starts to go wrong around Frame 24660:
> I did disable window scaling on Postfix
What do you mean by "on Postfix"? This is a kernel setting, modified
via "sysctl" or similar. In any case the session you posted has
no window scaling, so that is moot. You could try disabling
selective ACK support if that's possible:
net.ipv4.tcp_sack = 0
net.ipv4.tcp_dsack = 0
net.ipv4.tcp_fack = 0
> and for a while i thought it had
> solved the issue but then more of the same as you can see in the log.
> (hopefully showing the correct information!)
>
> TCPLog <http://postfix.1071664.n5.nabble.com/file/n57846/TCPLog>
Somebody's firewall or NAT device is pining for the scrap heap.
The receiver's duplicate acks eventually trigger sender RSTs instead
of retransmission of the lost segment. What we don't know is whose.
--
Viktor.
