Question about DHCP.

Glenn Satchell Glenn.Satchell at uniq.com.au
Tue Jul 8 14:24:07 UTC 2008


The time spent in recovering state is directly related to the MCLT
setting in the failover statement. This state is invoked when the two
dhcp servers decide that their leases files do not correspond. MCLT is
the duration of a lease when the dhcp server cnnot communicate with the
partner. So the two servers wait MCLT to make sure that any leases of
MCLT duration have expired. This is not a typical situation, but occurs
when, for example, the leases files are deleted.

Selecting a suitable value for MCLT can reduce the impact of this, but
the right answer is typically to not delete the leases file(s). The
lease state is stored in two locations: on each client, and on the dhcp
server. Deleting it in one place can cause problems, although typically
as each client tries to renew it's current lease the dhcp server will
re-assign the same IP address and things will generally get back in
step within one default lease period, but that might be a day or more
with typical setups.

The better solution is to add more IP addresses to the range if
possible. If you have a large laptop population then you might want to
give them shorter lease duration, so that a laptop that connects for
only a short period, say an hour or so, doesn't hold a dhcp lease for a
day or more.

The dhcp man pages provide good coverage of this topic, see the
dhcpd.conf man page and read the sections with 'FAILOVER' in their
sub-heading.

regards,
-glenn

>From: Eric Kenny <eric.kenny at marist.edu>
>Subject: Re: Question about DHCP.
>Date: Tue, 8 Jul 2008 08:52:37 -0400
>To: dhcp-users at isc.org
>
>We experienced the same behavior with our failover setup.  We have  
>LOTS of subnet declarations (most of them /21) and found that it  
>would take between 1.5 and 3.5 hours for the DHCP servers to come out  
>of "recovering" state.  As most would expect, this pretty much makes  
>failover useless to us.
>
>We have been looking into other options, like using hearbeat and  
>direct all DHCP requests to a "service IP" instead of the actual IPs  
>of the DHCP servers.
>
>If anyone has any better ideas, they would be greatly appreciated.
>
>> When I restarted the servers,
>> the both went into a "not responding (recovering)" for all the  
>> networks
>> that I have defined in my configuration file.  I thought that it would
>> just take a few minutes for the recovery process but I waited over an
>> hour and it was still not offering any leases.
>>
>> I had to delete the failover configuration and start up one of the two
>> servers to get the people to work.
>
>
>Eric J. Kenny
>Network Analyst
>Marist College
>3399 North Rd.
>Poughkeepsie, NY 12601
>845.575.3820
>



More information about the dhcp-users mailing list