ignored TTLs?

[aklist_bind at enigmedia.com]

Hi All:

I needed to change the IP address for a domain, whose refresh had been 
previously set to 3h (10800).

I set the TTL for the domain to 5m (300), and I left it there for 3 days.

I then changed the IP address for the corresponding A records (they did not 
have their own TTLs separate from the main TTL), set the TTL back to 3h, and 
reloaded Bind.

The changes of course were picked up immediately by my slaves and my 
upstream NS, and I thought all was fine. From my office, which is on a 
static DSL connection across the country, the domain resolved to the correct 
IP address.

Then I went home for the night...

When I checked mail in the morning, I had people around the country claiming 
they couldn't resolve the domain. I remoted into my home computer which is 
on a cable connection, and the proxy server it was connecting through was 
caching the old IP address. Some people being CC'd on emails said they had 
no problem resolving the domain, but I see traffic on the site (it receives 
about 5000 visits per day) was down significantly, so it looks like a lot of 
hardware that was caching NS didn't pick up the change, even 48 hours after 
I made it.

Question is, what did I do wrong? What do I need to do in the future to 
prevent it from happening again?

Right now, I put a redirect on the old IP address to the new IP, but I 
shouldn't have to do that normally?

TIA, Andrew