On Thu, 18 May 2000 11:30:42 -0300 (BRT),
Christian Robottom Reis <[EMAIL PROTECTED]> wrote:
>I remain unable to understand why my gateway is able to connect fine, and
>why _it_ doesn't defrag them so we can receive them internally. Changing
>MTU on the ppp link fixed the problem.
The path MTU is based on the MTU of the end points. So if an internal
machine starts a connection the path MTU is copied from the Ethernet,
i.e. 1500. If your gateway starts a connection the path MTU is taken
from the PPP link, say 576.
* For an internal machine the remote site sends a packet of 1500 bytes
marked do not fragment.
* That packet hits the PPP link which only handles 576.
* The ISP end of the link drops the packet, sending back ICMP "need to
fragment" to the server.
* Something along the way (probably a misconfigured firewall at the
server) drops the ICMP response so the server does not know about the
bottleneck and keeps sending oversized packets.
When the gateway starts a connection, it does so with a low path MTU so
the problem does not arise.
Not much you can do about this except log a problem with the servers.
-
To unsubscribe from this list: send the line "unsubscribe linux-net" in
the body of a message to [EMAIL PROTECTED]
