|
Any they should, since SE is gaining momentum on the market. For some
reason vodafone is busier figuring out how their SOA portfolio should
look like for their potential customers than to actually address
unimportant things like a buggy TCP stack. Either we find out directly
what's wrong with those phones or we try to find an elegant solution
for LVS, without breaking it for all the other people. The immediate
fix I have in mind is too intrusive and severely lacks security, so we
have to think about it a bit.
I found the problem!
Congratulations!
By solving another problem it appeared that new Sony Ericsson phones
cannot
reconstruct fragmented packets!
This is sad.
Our IPVS director host communicates with phones via GRE tunnel with MTU
1476.
So when full size packet arrives, director chops it into two fragments
sends
them to phone and SE phones goes crazy.
Can you do some more tests to verify this 100%? If this is the case,
I'll might get my August phone bill canceled, since I can claim that I
had a high re-transmission rate and connection lookup which caused the
huge amount of transferred bytes over edge yielding those costs which
had nothing to do with my field tests :).
To resolve it I had to add one more hack:
iptables -A INPUT -p tcp -m mark --mark 0x1 -m tcp --tcp-flags SYN,RST
SYN -m tcpmss --mss ! 0:1430 -j TCPMSS --set-mss 1430
Why don't you add this to your routing table 100?
ip route change default via $dgw dev $intf mtu 1430 table 100
Like this only GRE will be forced to a lower mtu and not the whole machine.
Now I'm changing MSS to 1430 from default 1460 and TCP/IP packets
arrives less
than 1476 bytes (tunnel interface MTU).
Best regards,
Roberto Nibali, ratz
--
echo
'[q]sa[ln0=aln256%Pln256/snlbx]sb3135071790101768542287578439snlbxq' | dc
|
