1

Specifically regarding ISC dhcpd (but information regarding the principle in general), what are reasons for choosing a fail-over configuration instead of each server having their own range?

Assuming you have more IPs to allocate than necessary, it seems the latter has much more flexibility as you can have more than two servers. Would the only reason be when you have less IPs to allocate than potential leases?

Or have I completely overlooked something...

Samuel Harmer
  • 214
  • 2
  • 8
  • Some things are connectionful, and those won't like changing IP address in the middle of a transaction. – MadHatter Sep 12 '16 at 08:29
  • Surely if you were running an IP change sensitive application you'd use static IPs or `fixed-address`es? I suppose what I'm trying to say is, what kind of application would be running on a host which relies on DHCP but is 'connectionful'? – Samuel Harmer Sep 12 '16 at 08:57
  • Some people prefer to run fully-dynamically-addressed networks in order to ease the pain of renumbering. This is particularly an issue with ipv6-enabled networks, where NAT is very frowned upon (ie, real addresses should be used all the way to the edge of the network) but PI space doesn't really exist, so a change in connectivity provider means a complete readdressing. – MadHatter Sep 12 '16 at 09:05

1 Answers1

1

ISC's DHCP Failover has some positive aspects, however, the overall configuration (from our internal testing) is less than optimal. The positives - especially for large organizations - include:

  • tracking of your entire IP Address Pool across all DHCP servers
  • a standard network router/switch configuration across your entire enterprise for DHCP helper addresses (assumes a segmented broadcast domain in IP4)
  • some very basic load balancing of DHCP traffic
  • no more single points of failure for DHCP in your environment

From our testing, though, it takes a fairly long amount of time for the server failover to actually happen and ties up some of your IP resources even after failover has occurred. So, some negatives:

  • failover is still something of a manual process (mitigated somewhat in releases greater than 4.2)
  • IP Pool is split across two servers
  • increased configuration complexity

As far as your scenario (having fewer IPs than potential leases) this wouldn't help, as each system needs to reserve part of the pool for itself, which could lead to IP resource contention.

Thomas N
  • 436
  • 2
  • 9