Q. If I'm using multi-site clustering with multiple IP addresses, how does DNS resolution work?

A. Normally, the active IP address (the IP address for the subnet where the resource is hosted) is registered in DNS with the network name. This allows clients to resolve the name to the active IP address. Now imagine a site failover occurs and a new IP address is registered with DNS for the network name. The problem is twofold.

Firstly, clients cache DNS entries for a period of time, the Time to Live (TTL) of the DNS record, before checking with DNS if the resolution has changed. Secondly, DNS is usually implemented as multimaster with Active Directory (AD) integration, so while an update to DNS may be made in one location it will take time for that DNS update to replicate to other DNS servers in different AD sites.

The first step you can take is to reduce the TTL of the DNS host record registered by the cluster service from the default value of 1200. For example, the Exchange team recommends a value of 300 (five minutes). This means clients would, at worst, keep trying the wrong IP address for five minutes before they check with DNS to see if the resolution address has changed.

The other step you can perform is to tell the cluster service to register all IP addresses for the network name, even if they’re not online. When the clients query DNS, they’ll be sent all possible IP address values. This is only an option if the application that is attempting to connect is smart enough to try all the returned IP addresses and not give up if no communication is possible on the first address. This means if the primary goes offline and moves to a new IP address, the client application already has the other IP address to try.

To change the TTL of the host record, use the command

Get-ClusterResource <Resource> | Set-ClusterParameter HostRecordTTL <new value>

To configure to register all IP addresses with the host record, use the command

Get-ClusterResource <Resource> | Set-ClusterParameter RegisterAllProvidersIP 1