dhcp failover

[Simon Hobson]

John Abbott wrote:

>Hi listers I am new to the list, in my experience this would only be a
>real issue when dhcp must server many hundreds+ of clients as each
>client when starting will get NAK and then start the negotiation from
>begining.

No, that is not what will happen !

Clients will not get a NAK unless the address they are currently 
using has been re-allocated (leased) to another device. If the client 
has been switched off since before the servers were switched then 
this won't matter, but if it's been on-line then it will have been 
running (or rather not) with a duplicate address.

>It has also all sorts of additional issues such as dups as
>mentioned for networks that don't respond quick enough,

It's nothing to do with 'networks that respond quickly enough', it's 
to do with whether devices respond to pings or not - given the number 
of systems that now have built-in firewalls that block pings (which I 
personally think is a stupid idea, but then that fits with my views 
on certain OSs anyway) there is a fair chance of something using an 
address but being invisible to the server.

What may happen is a storm of discover-offer-decline sequences as 
devices are offered a duplicate address and then detect (via ARP) 
that it's in use and so reject it. The server will then work through 
it's address space offering addresses until the client is happy.

>on large networks it can cause backoff on the switch port your server is
>connected to.

Can you elaborate on that ? I can't see why dhcp traffic would cause 
problems on a switch unless you've set some silly filtering/traffic 
shaping rules.

Simon