Hi!
I've set up an LVS, with the balancer linux 2.2.18 patched for lvs, and
some real servers
in direct routing.
I've tested real servers such as windows NT (too bad and lame but it runs
too in the cluster),
BSD and linux.
I'm in trouble with real servers running linux kernel 2.4.
They respond to arp requests in spite of the linux director, in which way
can I stop them from responding to arp requests?
I've configured them in this way:
echo 1 > /proc/sys/net/ipv4/ip_forward
ifconfig eth0 192.168.0.10 netmask 192.168.0.128 broadcast 192.168.0.127 up
ifconfig lo:0 192.168.0.5 netmask 255.255.255.255 broadcast 192.168.0.5 up
route add -host 192.168.0.5 dev lo:0
Here I've used 192.168 ips as an example in spite of public ips.
In the example, ip 192.168.0.5 is the ip of the cluster, which is also on
the eth0:0 of the real server.
If I turn off the linux 2.4 boxes everything is good and from the LVS I can
see connections
distributed correctly.
If I turn on the linux 2.4 boxes, no connections are managed by the lvs,
and if I try to telnet to the 192.168.0.5 ip, I login in the linux 2.4 real
server, which is responding to all arp requests for that ip.
I hope you can help me or suggest what to do.
|