|
>How about using clusterip to balance across two LVS balancers? That would
>be the best of both worlds. Still centralized management, but instead of
>failover you just have two boxes already working. I don't know if the
hash
>table can be changed on the fly.. but this might be more effective than
>gratuitous arp like heartbeat does.
>This sounds a lot like CARP, but done at the IP level.
_______________________________________________
You need to be extremely careful when considering something like this.
There are many possible theories that are not at all practical.
Hot spare serves as exactly that. This is what LVS (and HA in general)
lives for. It is a spare. Meaning that it can fall in to "Replace" a
production server with little/no manpower. Operative word being "Replace".
When considering going live with both LVS servers, you leave vulnerability
if one ever goes down. This vulnerability is lack of capacity if the
servers aren't serious overkill to begin with (which rectifies the need to
balance them anyways, don't it?).
One goes down, the other has to take the load of 2 servers.
P
|
