mac adress problem ?

Hi,

I use a temporary load balancer, LBtemp, redhat 7.3, kernel 2.4.18-14, 
IPVS 1.0.4 (source) and piranha 0.7.0-3 (source). No hidden patch on 
this LB.

All my web servers are redhat 7.2, kernel 2.4.19 with hidden patch 
(hidden-2.4.19pre5-1.diff).

Everything is working fine with my current setup. I use direct routing.

Right now, I setup two new load balancer LB1 and LB2. One primary and on 
secondary : redhat 8.0, kernel 2.4.19 with hidden patch 
(hidden-2.4.19pre5-1.diff) and ipvs-patch 
(linux-2.4.19-ipvs-1.0.7.patch), ipvsadm-1.21 (source) and 
piranha-0.7.0-3 (source).

So, here is my problem....

If LB temp is working and I stop pulse on it, then  start pulse on LB1 
in order to swap on it, the load balancing is working fine.

If LB temp is working and I stop pulse on it, then  start pulse on LB2 
in order to swap on it, the load balancing doesn't work

If LB1 is working and I stop pulse on it, then  start pulse on LB2 in 
order to swap on it, the load balancing doesn't work (so the failover 
will not work if my primary fail).

I think I found the problem on LB2. If I changed the ip of all the 
virtual server (in lvs.conf for piranha on LB2 and the loop back adress 
on all my web server) LB2 is working fine !!!!

So the problem is due to the mac adress ?? But how can you explain that 
I'm able to swith from to LBtemp to LB1 (no mac adress problem) but not 
from LBtemp to LB2 (mac adress problem ?) ??

Thanks for your help.

Cyrille.