Failover Clarification

[Glenn Satchell]

>From: "Benjamin Wiechman" <benw at meltel.com>
>To: <dhcp-users at isc.org>
>Subject: RE: Failover Clarification
>Date: Wed, 9 May 2007 08:44:43 -0500
>
>> The two servers communicate with each other frequently, and go through
>> various pool balancing exercises so that they tend to keep an equal
>> number of free leases on each server.
>
>
>So in theory if I have a couple of free leases it *should* work. In practice
>how often do the servers work to balance the leases? I'm not seeing a
>configuration directive that would define this time. I was reading that if
>the free address are within about 10% it the peers don't really work to
>balance the addresses, is this still true?

The whole balancing algorithm, changed in 3.1.0a (alpha). I'm running
3.1.0a3 and on an idle network I see a balancing log entry every 60
minutes. This period is completely configurable in dhcpd.conf. Quickly
looking at the dhcpd.conf man pages between 3.0.5 and 3.0.0a3 I think
you definitely want 3.1.0 for your situation.

>> When you're down to the last free IP one server will not reply because
>> it has no lease to offer. The other server should offer the available
>> lease that it has.
>
>OK. That is reasonable. So if I'm running a dhcp relay through my switches I
>would need to be doing a directed broadcast to allow the servers to perform
>their own load balancing, rather than a relay by IP.

I've always configured two relay statements so that requests were
forwarded to both dhcp servers.

>> About a year ago I did some work for a client to set up a script which
>> ran on each dhcp server. When communication was lost for a certain
>> amount of time (30 or 60 minutes I think) it used the omshell interface
>> to switch the server to partner down. When the partner communications
>> returns dhcpd automatically return to failover mode by itself.
>
>> The details should be in the archives. It was a pretty simple script,
>> and it only got used once in a DR test where we manually switched off
>> the network between sites. This client had redundant networks, etc, so
>> we never had a real failure to test it out.
>>
>
>Thanks for the pointer. I did see that yesterday and have been playing with
>it. I was just curious if anyone had tried this in a similar circumstance.
>Most people I speak with don't have a similar configuration.

regards,
-glenn