DHCP Relay agent not forwarding messages to the client

[Graham Clinch]

Hi Folks,

On 16/06/2015 14:28, Friesen, Don MTIC:EX wrote:
> Yes, this is the expected and normal behavior.  Each server will reply 
> with an address from its half of the pool. The machine that is receiving 
> the offers will take the first offer to arrive and ignore the other.

We have 'split 128' & 'load balance max seconds 5' parameters set, and so clients that discover with a low 'secs' field are only answered by one of the pair:

==
Jun 16 09:19:05 is-dhcp0 dhcpd: DHCPDISCOVER from aa:bb:cc:dd:ee:ff (Kates-iPhone) via 10.32.176.8: load balance to peer schlep-failover
[and no OFFER]

Jun 16 09:19:05 fa-dhcp0 dhcpd: DHCPDISCOVER from aa:bb:cc:dd:ee:ff (Kates-iPhone) via 10.32.176.8
Jun 16 09:19:06 fa-dhcp0 dhcpd: DHCPOFFER on 10.32.186.98 to aa:bb:cc:dd:ee:ff (Kates-iPhone) via 10.32.176.8
==


Some clients don't implement the secs field correctly (or appear to get their most/least significant bits muddled), so we do see some clients that receive two offers.  Perhaps there are some relay agents that are the same (although I'd be surprised if the ISC relay was broken in this regard).

The idea here is that, on average, the load for unicast renews is split evenly between the pair of dhcp servers.


On 16/06/2015 20:16, Sean McMurray wrote:
> If they are both making offers, what is the point of configuring them in 
> failover? You might as well set them up independent of each other and 
> have them serve different pools.

Even if all clients receive two offers, the benefit is that when one of the dhcp pair fails, clients don't suddenly renumber themselves into the remaining server's pool - they can continue to renew their current address (and that you don't need to run with two pools, each large enough to cover all clients).

We're very happy with relay agents and failover - 1,150 networks, 17,300 static 'host' addresses & 260,339 dynamic addresses.

Graham

PS: Any movements on failover for DHCPv6?  At least I don't feel bad for using twice the address space within a /64.. :)