Advantages of DNS-based failover:
- Flexible in placement of nodes since they can be at any provider.
Some of those advantages probably should be merged depending on your
point of view.
Disadvantages of DNS-based failover:
- Comparatively slow (minutes) even in the ideal case.
- Some client software (e.g. web browsers) may be reluctant to
update IP addresses they have cached, meaning the downtime is not
in your control.
- Some ISP resolvers enforce a minimum TTL meaning they will ignore
your failover name's lower TTL and cache the record for as long as
they are configured to. Some claim that AOL have it set to a week
but I haven't seen any proof of this. Here are some anecdotes:
My experience would tend to agree that you could expect failover
time of a couple of minutes for the vast majority of clients but a
minority of clients won't see the new IP for days.
- Having to set a low TTL (e.g. 300 seconds) on one of your host
names means that your own DNS infrastructure is going to receive a
lot more traffic. DNS outages will hit you harder because they
will be noticed within your low TTL for almost all clients.
Normally you would do layer 2 high availability (moving an IP
address around) inside one datacentre and then might consider
DNS-based load balancing/HA between multiple sites.
There's also anycast which is a bit beyond the scope of this
conversation and isn't something BitFolk will be able to provide you
with in the near future.
On Wed, Jan 19, 2011 at 01:40:51AM +0000, Andy Loughran wrote:
What advantage does Up takeover have over a 5 minute ttl on dns?
Aside from the obvious, in that ip change can have s lower change time.