|
On Sat, 17 Mar 2007, Hideki wrote:
Hi.
I have currently set up LVS like this,
| WAN : Internet | - | PPPoE Router + LVS | - | LAN : 2 real servers |
The PPPoE router is a single machine (for now) which handles the PPPoE
connection as well as LVS (in NAT form with least connection forwarding
method) toward its real servers (HTTP) in the LAN. When I access the LVS from
inside the LAN, the requests get passed to the real servers and back to the
client fine. But when accessed from the WAN, the machines stop responding
after the content size goes beyond 1402 bytes. I just guessed this has
something to do with PPPoE packet size and that is the reason server stops
responding right after the size becomes 1403 bytes big.
hmm. icmp need_defrag packets are not getting back to the
internet side of the PPPoE segment. They may not be being
generated. It would help to debug the problem to know what's
going on in this regard. Can you run tcpdump on the outside
of the director on a client connection with < 1402 and >1402
bytes and see if the icmp packet is generated? If it is, can
you figure out why the icmp packet is not making it to the
client?
Noone has ever mentioned using PPPoE with LVS before, so I
can't come out and say that we know it works. (I would have
hoped it did, but that's not the same thing.)
What happens if you just have a bare realserver on the
server end (ie rather than doing through the director)?
PPPoE must be a solved problem already - presumably it's
working everywhere else for packets > 1402.
I have a 2.6.19 kernel and version 1.24 of ipvsadm.
In the iptables on the router, I have the mss fix packet mangling, so that
SNAT works fine fetching data from the internet back to the LAN machines.
iptables -t mangle -A POSTROUTING -p tcp --tcp-flags SYN,RST SYN -o ppp0 -j
TCPMSS --clamp-mss-to-pmtu
does this help or do nothing? If PMTU is working, you
shouldn't need this (clamping mss to the pmtu size sounds
like a no-op to me). If PMTU is not working, then you'll
need something like the command in the HOWTO section on
LVS-Tun to handle the reduced payload for ipip packets. If
that fixes it, then we've got another bug.
Joe
--
Joseph Mack NA3T EME(B,D), FM05lw North Carolina
jmack (at) wm7d (dot) net - azimuthal equidistant map
generator at http://www.wm7d.net/azproj.shtml
Homepage http://www.austintek.com/ It's GNU/Linux!
|
