Failover Clarification

[benw at meltel.com]

We are looking at migrating to a dhcp failover configuration and I was
hoping for a bit of clarification.
 

When using failover the servers perform load balancing. This is reason for
the split directive from what I can tell. In essence each server should
handle approximately 50% of the load. Correct?

 

We are a wireless ISP. To help control broadcast storms, which would cripple
us due to the relatively low pps capabilities of our wireless equipment we
run rather small subnets on our access points, typically a /27 network per
AP, adding additional subnets as required. However to keep our utilization
up so we can get more IPs as we need them we run the ranges up to 1 free
address. In a failover environment this seems like it is going to cause some
problems. When there is the state that one server has one address free, and
the other has zero, if a host requests an IP address (say for argument that
it hasn't had an IP before - a new NIC, etc) from the server with zero
available addresses does that server go to the peer to determine if there
really are no free leases, or does it simply respond that it doesn't have
any free leases?

 

Does anyone have some experience running a failover configuration with
address pools that are this full? If we keep a pair of free leases available
is that going to be sufficient in most cases? Our current lease time is 8
hours. This could be lowered to more quickly return addresses to the pool if
required, but at some point I would think this is going to start to work
against us as we reach 2000-3000 subscribers. 

 

There is also the question of putting one peer in a partner down state and
how soon this would have to be done.

 

Any comments from anyone in a similar situation would be greatly
appreciated. 

 

Ben Wiechman

ben at wisper-wireless.com