Failover on a large DHCP system
cra at WPI.EDU
Fri May 8 20:45:04 UTC 2009
On Fri, May 08, 2009 at 11:00:33AM -0600, Nicholas F Miller wrote:
>> On Fri, May 08, 2009 at 10:09:29AM -0600, Nicholas F Miller wrote:
>> > With constant restarts of DHCP and the need for
>> > failover to redistribute the pools on restarts I think the server
>> load and
>> > disk access would be too high.
>> The disk access footprint between failover restarts is only the
>> "changes" to the database. It doesn't remirror the lease db every
>> reconnect - they only update each other with any new bindings that
>> ocurred during the blackout.
>> With huge numbers of leases, it can be a problem however, as it causes
>> both servers to traverse the entire lease database (in memory) right
> With around 72,000 leases at any given time and 677 pools, should we
> just continue using our current restart methodology or try to implement
> omshell as a way to add/remove the hosts? Since omshell writes its
> changes to the leases file, the changes will survive a restart, ignoring
> the config. This could lead to some interesting troubleshooting if
> something were to go awry.
We are using failover and restarting the server at most every 20
minutes for updates from a database-backed registration system. We
have about 120 pools and about 10,000 addresses in the pools (but not
very full utlization of that overall). The overwhelming majority of
requests match static fixed-addresses.
More information about the dhcp-users