|
Jacob Coby wrote:
>
> > I don't know what happens here, but I thought the first machine that
> > couldn't handle the large packet would frag it and pass it along.
>
> Really? That would mean a bug in LVS?
Julian, Ratz,
Jacob is having problems when doing an http POST of a large
file (>934 bytes, I believe) from a windows client to a single
director, 2 realserver LVS-Tun. The connection hangs and the
client doesn't receive a reply. The problem only occurs when
the windows client is on the same network as the director.
Clients outside the local network don't have problems.
He hasn't tested with a linux client on the local network.
tcpdumps of various machines, show that the client gets a
"need to frag" icmp response.
The windows client appears to then send a frag'ed set of packets
> it appears so, yes. I'm seeing things like
> P 632:2092(1460) ack 1 win 17520 (DF)
> which would indicate that the client is downsizing the packet to 1460 (1480
> with the IP header).
He is going to check the
/proc/sys/net/ipv4/ipfrag_*
settings on the director as suggested by Sebastien Bonnet.
Is it possible that the director has problems with large packets?
I don't know which kernel/ipvs he's using.
Joe
--
Joseph Mack PhD, Senior Systems Engineer, SAIC contractor
to the National Environmental Supercomputer Center,
ph# 919-541-0007, RTP, NC, USA. mailto:mack.joseph@xxxxxxx
|
