>The asymmetry of this seems quite bizarre.
I think we´re getting really close now:
1. If I request a file from realserver, the realserver sends big data
packets and the client is sending small control packets back.
In this case, the realserver should receive a NEED TO FRAGMENT ICMP packet
but doesn´t get it.
2. If I send a file to realserver, the client sends big data packets and the
realserver is sending small control packets back.
In this case the client _receives_ (seen by tcpdump) an ICMP DEST
UNREACHABLE NEED TO FRAGMENT from it´s default router and sets down the mss.
IMHO the question is, why doesn´t the RS receives any ICMP need to frag from
director2.
In case of an iptables-only connection (SNAT/DNAT) without ipvsadm
everything works fine.
>I did spend some time trying to reproduce this problem using 2.6.27-rc1
with no success (by which I mean pmtu worked fine and I did not see the
problem you see). Are you still using the etch kernel?
>My testing involved setting the ttl on the tunnel (to 64 from memory) and
not using nopmtudisc.
>I will try again with something much closer to the setting that you posted
in a different message[1]
>[1]
http://archive.linuxvirtualserver.org/html/lvs-users/2008-07/msg00110.html
I´m still using 2.4.18 from etch. We´re trying to update to etchnhalf this
week.
We tried also playing with different ttl/inherit, pmtudisc/nopmtudisc with
no success.
cheers
Marco
|