[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: problems with having two DHCP servers...

John A. Sullivan III wrote:
> The problem is when they are administering addresses in the same range.
> I've not configured DHCP for a long time so maybe this is common now and
> the problems have all been resolved

I believe the problem is sufficiently resolved now.  Time has past and
this is definitely something where improved code has improved the
situation quite a bit.

>                                     but, in the past, if one wanted
> redundancy, one would administer different ranges on the same subnet so
> that there would be no conflicts.

I think the best in class solution available today is to use the ISC
DHCP daemon.  This is packaged for Debian as isc-dhcp-server.  Then
configure two machines in a failover configuration.  This works very
well.  I am doing this on a variety of sites.  It is very robust and
works great.

Sometimes people curse it though due to a self-inflicted configuration
problem.  I have seen people try this repeatedly for some reason.  And
I have a hard time understanding why.  They try to use half as many IP
addresses in the pool as they need.  I don't know why but they do.
There are a *lot* of addresses available from the RFC1918 private
ranges and even on large networks I have a hard time conceiving of
being short of IP addresses.

But people seem to often configure only half as many addresses as they
need.  Then when one server is offline the remaining server, which is
working perfectly as configured, runs out of IP addresses to assign
because the pool is only half as large as it should be in order to
support the network.

I think people hit this case because when both servers are up and
online they effectively have load sharing and so each has 50% of their
pool free.  So they use a pool that is too small because they are
fooled into thinking of the case with both servers running instead of
thinking of the case of a server being disconnected.  But it is the
disconnected case that is the important one!  The simple solution is
very simple.  Use a large enough pool and this problem is avoided.


Attachment: signature.asc
Description: Digital signature

Reply to: