Re: dns round robin - use for redundancy

["John Madden" <jmadden@ivy.tec.in.us>]

> Doesn't DNS round-robin usually just rotate the entire list of
> addresses that it returns?  So if ldap.foo.bar resolves to addresses 1,
> 2, and 3, one query would get 1,2,3; the next 2,3,1; the next 3,1,2;
> etc.  If 2 goes down, the second query would result in first timing out
> while connecting to 2, then trying 3 and presumably succeeding, no?
> (I'm assuming that OpenLDAP follows the best-practice of trying each
> returned address in order until a connection is established or the list
> is consumed.)

Only applicable if the application knows to "try again" if a connection
times out or a 'connection refused' is returned.  Also, depending on the
TTL, what DNS is returning could result in a failure for the length of that
TTL, and for the length of that TTL every time through the round robin.  Any
way you look at it, for a time-critical application, DNS should be
dynamically updated.

...And that's where things like load balancers and virtual servers and such
come in.  For example, ldap.yourdomain.com points to a box that does nothing
but port forward with balancing between a pool of servers and pull them out
automatically if it isn't able to connect to them itself.

However, DNS round robin is a nice 'cheap' solution.


John





-- 
John Madden
UNIX Systems Engineer
Ivy Tech State College
jmadden@ivy.tec.in.us