I have to say I'm having a similar issue with load balancing. This morning is the first time I have really played with it. Here is my setup ina nutshell :-
2 firewalls in a carp cluster, static public IPs on the WAN, static public IPs on DMZ, and Private IPs on LAN.
NAT is only used for the LAN -> WAN connection. The DMZ servers can route to special ports / IPs on the LAN
I set up a pool containing 2 LAN IPs and setup a virtual server on the LAN carp address. The DMZ connect to the virtual server to process some fast-cgi stuff. I added a rule on the DMZ interface to use the POOL as the gateway as suggested by GruensFroeschli (although this feels more like its for outbound load balancing of WAN connections, not inbound server balancing?).
Sticky connections is off.
The server always connects to one backend server (POOL is set to load balance, not failover). If I stop the service on the LAN IP thats getting all the connections, the first couple of connections fail, then they start going to the second LAN IP.
After starting the service on the first LAN IP again, the next connection continues with the second LAN IP, then after that all connections revert back to the first LAN IP again.
I'd like to :-
a) Have it share the connections round robin style against the two LAN IPs
b) When one does go down, have all connections seamlessly go directly to the second, not have a couple of failures like I see at the moment.
Is this possible?
Regards
Ben