Using a low TTL to enable a fail-over cluster?

Pete Ehlke pde at ehlke.net
Fri Mar 22 16:40:45 UTC 2002


On Fri, Mar 22, 2002 at 05:00:36PM +0100, Jakob Bak wrote:
> 
> The idea is then, that in case our primary host is down all traffic will go
> to the second cluster as the secondary nameserver takes over for the
> unavailable primary. This is of course pointless if our DNS information is
> cached all over the internet for e.g. 24 hours, so it is essential that we
> can set TTL to for example 5 minutes and that at least a majority of DNS
> servers respect this TTL. Secondly, we will have to scale our secondary
> cluster according to the load which it will receive (from the secondary DNS
> server).
> 
DNS based load balancing and/or failover is a touchy subject. There are
a fair number of companies that have tried, with varying levels of
success, to build and deploy products and services based on some more or
less sophisticated use of the concept, and while some have been fairly
successful from a business perspective, I'd have to side with those
greater wizards than I who opine that it's always an abuse of the
protocol and rarely what you really want to do.

Search the NANOG archives for this stuff. There have been several rather
lively discussions of thi issue over the past few years. The last I
recall was in October of 2001, under the thread 'dns based
loadbalancing/failover'.

-Pete
-- 
"religious fanatics are not part of my desired user base." 
- djb at cr.yp.to


More information about the bind-users mailing list