|
> Seriously, I think the thing we might be doing that's different from
> what most people are using this for is that we have about 50-60 aliased
> VIPs on the outside that each map to a pair of servers inside, one
> primary and one backup. We're using it for failover on the internal
> servers rather than real load balancing for a single external VIP
> against many internal servers. We also have a bunch of portfw rules set
> up for 1:1 access from external aliases to internal addresses on the
> primary and backup servers.
I have a feeling that it has something to do with the NAT rather than LVS.
I use LVS to load balance about 1024 VIPs onto 4 real servers using VS-DR. I
have not had any problems with only 4096 file handles. I'm very sure that I
doubt I use even 1000 file handles under normal operation.
I don't use RedHat's HA package either, however, which could be the source
of the problem. I haven't really looked at any of the packages included in
RH's compilation.
Could you include an example output from an ipvsadm -L ? About how many
active connections do you normally get to each server? About how many of
active connections total using the aggregate of all of the servers? How
about inactive connections?
Now I usually don't do much kernel or filesystem development, but the inode
problem you mention makes me think even more that it isn't an LVS problem.
That's just my two cents. From your description of what you're doing, I
really don't think that you're doing anything too exotic.
Perhaps you could send us some of your configuration files? Perhaps the
output of some of the RedHat-HA utilities? Give us a better model of your
system. Maybe someone will come up with some other ideas.
All the best --
Ted
|
