dhcp server on cluster best practices

Glenn Satchell glenn.satchell at uniq.com.au
Tue Oct 18 11:48:39 UTC 2011


Hi José

dhcpd 3.0.3 is *very* old and has a lot of known bugs with failover. You 
will do yourself a big favour if you upgrade to a newer version. 3.1 
would be the minimum, but either of the latest 4.1 or 4.2 releases are 
also known to be good. Your config and lease files should work without 
change on the new versions.

regards,
-glenn

On 10/18/11 05:43, Simon Hobson wrote:
> José Queiroz wrote:
>
>> I configured these machines to work on failover mode, so each cluster
>> node is a failover peer.
>>
>> The problem is that now we'll have a third cluster node, and the
>> failover mode will no more be feasible.
>
> Ok, I see a number of options.
> You could continue with just two nodes running DHCP - losing one would
> be irritating, losing a second could be considered careless ;-)
>
> Or, you could split the pools up and run a three way failover - whilst
> any pool may only be shared across two servers, you can share different
> pools with different servers. So setup three pools (a, b, and c), and
> have three servers (1, 2, and 3).
>
> Pool a has failover between 1 & 2
> Pool b has failover between 2 & 3
> Pool c has failover between 1 & 3
>
> So each server supports 2/3 of your address space, with failover to two
> separate servers.
> If you lose any one server then "no-one notices"
> If you lose two servers, then only 1/3 of the address space is affected,
> and clients will migrate to the other servers as their leases expire.
>
> Of course, load is also plit 3 ways.
>



More information about the dhcp-users mailing list