Sonic:
> We're getting many lost connections from our new phone systems
> voicemail to email service. The actual voicemails and other emails
> send OK but we also see constant hits anywhere from every minute to 8
> minutes on the mail server from the phone system that are lost
> connections. The phone system is an NEC SV9100. Sender and Postfix are
> on the same local vlan.
> A snippet from the logs:
> ===============================================
> Sep 11 09:41:46 example00 postfix/smtpd[10259]: connect from
> sv9100.example.com[10.6.600.11]
> Sep 11 09:41:46 example00 postfix/smtpd[10259]: 8AF0A403E24:
> client=sv9100.example.com[10.6.600.11]
> Sep 11 09:41:46 example00 postfix/cleanup[10262]: 8AF0A403E24: message-id=<>
> Sep 11 09:42:03 example00 postfix/smtpd[10259]: lost connection after
> DATA (66836 bytes) from sv9100.example.com[10.6.600.11]
Also:
> DATA (66621 bytes) from sv9100.example.com[10.6.600.11]
> DATA (66767 bytes) from sv9100.example.com[10.6.600.11]
> DATA (66828 bytes) from sv9100.example.com[10.6.600.11]
Note that:
66836 = 65536 + 1300
66621 = 65536 + 1085
66767 = 65536 + 1231
66828 = 65536 + 1292
All of these are less than 65536 plus 1460 (one ethernet TCP payload).
You need to find someone who can debug TCP-level problems
with a network skiffer. I do not have time for that anymore.
Random suggestions:
- Disable TCP selective ack (sack) in the kernel.
- Disable TCP window scaling (wscale) in the kernel.
Usually, network-level trouble is caused by in-between boxes
(firewalls, traffic shapers) that mis-handle TCP.
Wietse
