[NTLUG:Discuss] Use DNS for redundant geographic sites?

[ntlug at thorshammer.org]

> My ultimate goal is to create a web site that will
> always be available.
> 
> I have two datacenters in two different cities so 
> I thought I could use a server in each of them to
> give me ultimate redundancy.
> 
> Any suggestions?

`	What kind of redundancy are you looking for? Primary and flip a switch to backup, both active/load balanced, etc? How long after a failure is it acceptable to be up and running again: realtime, < 1min, < 5min, < 30min, etc? What kind of data needs to be mirrored and how often does it change: static content (low updates), cataloged content (freq updates), transactions (realtime updates)?

	An easy solution is to have your DNS server give out both IPs to both sites (assuming you need non-realtime sync'ing between locations). Each location monitors the other and stops giving it out when it detects a failure. In the event of a failure, there will be a 5 min window where 50% of your traffic will probably fail (hitting the down node). Your outage window will be the sum of your polling interval and the DNS cache timeout. If you poll every 30 seconds and the DNS cache timeout is 5 min, then the longest outage is 5.5 min.

	Can your two datacenters route each other's IPs (more than likely yes if they are the same company)? This will provide failover at the IP level and take DNS out of it. This can lower the failover window depending on what kind of outage it is (site down, IP unreachable, etc). This won't correct apache failing (or anything else above layer 4 of the OSI model).

Robert