22 Jan 09 DNS based Load Balancing

There are two main options for DNS based load balancing. The first and most simple is the round robin option. We can use this for ‘A (alias) records’ and ‘MX (Mail-eXchanger) records’.

We can specify a priority for MX records. If we specify the same priority for multiple MX records, the querying client will toss a coin and ‘randomly’ decide which to use. The same applies to A records. This should provide with a reasonable split between your various records however provides no mechanism for server loads or using any kind of intelligence to route queries.

Another option is to return a record based on intelligence. Assume we are trying to balance load between web servers. The two popular methods we can use are to return a record based on knowledge of the load of the web servers, or alternatively return a record based on originating IP (location) of the requesting client.

This is all well and good however there are a number of considerations, specifically that DNS was not intended to be operated in this way.

1. You can set your records expire time to as low as you like, it will still be cached in circumstances by the browser and/or the resolver. This method will not account for ‘downed’ or ‘overloaded’ servers, they will still receive traffic.
2. Due to caching, should your browser or resolver hold on to the record, it will blindly access the same IP next time the host name is requested, without requerying the DNS server and ignorant of the changed network conditions.

Tags: dns, load balance, round robin