A DHCP relay agent is a computer that relays DHCP and BOOTP messages between clients and servers on different subnets. This way, you can have a single DHCP server handle several subnets without the DHCP server being connected directly to those subnets.

There are two ways to provide fault tolerance. The first method to provide fault tolerance, you can split each of your scopes (e.g., apply an 80/20 rule) between two servers. For two subnets, for example configure each DHCP server with a superscope consisting of two scopes, one scope for subnet A (scope 1) and one scope for subnet B (scope 2). Make sure that the scopes for DHCP server A and DHCP server B do not overlap. Also make sure that the IP address ranges are split according with the 80/20 (or 75/25) rule, so that the local scope uses 80% (or similar) of the local IP address range, and the remote scope uses 20% (or similar) of the remote IP address range. Of course for this to work, your router has to support DHCP relay.

With the 80/20 rule only holds true unless it takes longer than a few hours to get your failed DHCP server working again before you run out of IP addresses. Therefore, some administrators prefer a more conservative 50/50 rule, which actually works even better in a single subnet environment anyway, provided both servers have enough addresses scoped to cover the needs of all the clients on the network. Then if one server goes down, the other one can take up the slack for as long as it takes to get the first one up and running again.

Another method for DHCP fault tolerance is to form an active-passive fail-over cluster. By using two servers, you can create an active-passive fail-over cluster. The active node will service the DHCP clients while the passive node waits until the active node fails. When the active node fails, the passive node takes over the DHCP resources and becomes the active node.