By 'DNS failover' I take it you mean DNS Round Robin combined avec certains monitoring, i.e. publishing multiple IP addresses for a DNS hostname, et removing a dead address quand monitoring detects that a server is down. Cela peut be workable for small, less trafficked websites.
By design, quand you answer a DNS request you aussi provide a Time To Live (TTL) for le response you hand out. In autre words, you're telling autre DNS servers et caches "you may store this answer et use it for x minutes avant checking back avec me". The drawbacks come depuis this:
With DNS failover, a unknown percentage of votre users will have votre DNS data cached avec varying amounts of TTL left. Until le TTL expires these may connect to le dead server. Il y a faster ways of completing failover than this.
Because of le above, you're inclined to définissez le TTL assez low, say 5-10 minutes. But setting it higher gives a (very small) performance benefit, et may help votre DNS propagation work reliably even si there is a short glitch in network traffic. So using DNS based failover goes against high TTLs, mais high TTLs are a part of DNS et can be useful.
The more common methods of getting good uptime involve:
Placing servers together on le même LAN.
Place le LAN in a datacenter avec highly disponible power et network planes.
Use a HTTP load balancer to spread load et fail over on individual server failures.
Get le level of redundancy / expected uptime you require for votre firewalls, load balancers et switches.
Have a communication strategy in place for full-datacenter failures, et le occasional failure of a switch / database server / autre resource that cannot easily be mirrored.
A très small minority of web sites use multi-datacenter setups, avec 'geo-balancing' entre datacenters.
