Round robin, failover and resolvers

[Barry Margolin]

In article <bmlvv4$vah$1 at sf1.isc.org>, Ori Tend <ori at fetchbook.info> wrote:
>Dear list, 
>I have 2 boxes, and I'm trying to implement standard failover between
>the boxes- ie, if a box fails, requests would be served from other box.
>I've seen a number of commercial hosted solutions that cost $300 / month
>(looks like they host something like lbnamed). Trying to cut costs, I've
>thought of acheiving that myself using bind:
>Have a standard round-robin of the 2 A ip's, with very low ttl, assuming
>that in case of a failure of one ip, the client's resolver will try to
>hit the other ip, which will work well - as suggested by many posts in
>the list.
>However, I keep seeing "ancient resolvers would only use the first ip on
>the list", and I'm trying to understand, how ancient?
>Would Win98 resolver try to use the next ip on the list? Would Win2k? 

I think it refers to 15-20 year old Unix resolvers, which used a structure
that only had a place for 1 IP address.

However, it's not just the resolver that you have to worry about.  The
resolver just returns a list of IP addresses, it's up to the application
what it does with them.  Many applications only use the first address in
the list.

If you want a real failover solution, you need to use something like
lbnamed or Distributed Director.

-- 
Barry Margolin, barry.margolin at level3.com
Level(3), Woburn, MA
*** DON'T SEND TECHNICAL QUESTIONS DIRECTLY TO ME, post them to newsgroups.
Please DON'T copy followups to me -- I'll assume it wasn't posted to the group.