Failover and config differences

Phil Mayers p.mayers at imperial.ac.uk
Mon Dec 17 09:42:49 UTC 2012


On 12/17/2012 06:37 AM, Bjarne Blichfeldt wrote:
> I think you two are talking about different things.

That's possible. I must admit I don't understand where the "30%" number 
comes from in Chris' reply, but I do treat ISC dhcpd as more of a black 
box than most of our services, so it is entirely possibly the 
misunderstanding is mine.

> The original poster , as I understand it, has a setup where the
> available IP space is divided in 2 adress ranges, each served by its
> own dhcp server. If one dhcp server drops out, there is only one
> server left and also only half the address space available. This is
> no longer going to work, hence  "one server is no longer enough to
> serve the number of clients". Not because of speed limits, but
> because of lack of IP addresses.

This is exactly correct.

> In a failover setup, the resulting server would be set to "partner
> down" and handle the entire address range. One dhcp server is now
> able to "serve the number of clients".

Also correct, with one addition - I believe the remaining server will be 
able to *extend* existing leases that were originally handed out by the 
dead server, for some length of time that is a function of the lease 
time and mclt?

Therefore, if I set the least time correctly, know our time-to-fix, 
*and* over-size the pool sufficiently that the "free" portion of the 
remaining server pool half can handle our mean client arrival rate, we 
can remain fairly safely in-service.

>
> We have the second setup and that works nicely - within the
> limitations stated in the documentation. The config files in our
> setup are generated from a management station and pushed to the dhcp
> servers which  are restarted after config change. There is usually
> about a gap of about 10 minutes where the config files differ. This
> has so far not been a problem.

So it sounds like my main question is answered by several people - minor 
config differences don't seem to affect failover. There is obviously 
some mechanism in the protocol which can handle these, though it's not 
obvious to me reading the RFC which bit that is.

Anyway - that's good to know and was really all I was looking for. 
Thanks all!

Regards,
Phil


More information about the dhcp-users mailing list