Dynamic DNS (address takeover-type-things)

Michael Rowan mtr at cutaway.com
Mon Oct 19 21:47:54 MDT 1998


alanr at bell-labs.com wrote:
> 
> I sent this more than 12 hours ago, and didn't see the return mail, so
> I'm sending it out again.  Sorry if you've seen it twice.  You can read
> about Dynamic DNS at:
>                 http://davidsimmons.com/tips/081797/
> -------------------------------------------------------------------------------
> Another thing I assume we would eventually like to do is to perform
> Dyanmic DNS
> reconfigurations on the basis of node availability.
> 
> Let me elaborate:
> In DNS one can have a round-robin scheme so that when a user requests a
> particular
> machine (like www.netscape.com), that it will be redirected to one of a
> set of
> machines in a round-robin fashion.  In Dynamic DNS, one can dynamically
> change the
> mapping of machine names to IP addresses.  Ultimately, if we add a
> machine to or
> remove a machine from a cluster served by this kind of DNS addressing,
> we would
> want to be able to update the DNS round-robin list as well.  For these
> kinds of
> applications, this would be much better than simply doing IP address
> takeover,
> since the load will remain balanced in this case.
> 
> For example, if I have three machines in a cluster, and one goes away,
> IP (or MAC)
> address takeover would give ALL the new load to one of the machines,
> making it
> twice as busy as the other.
> 
> DNS round-robin updates would allow the machines to share the load
> equally.
> I'm not 100% sure which particular DNS servers will allow you to change
> the
> configurations of round-robin groups.  If none do, then it certainly is
> a desirable
> feature.


Seems like a useful feature for web servers or, depending on
the setup, OPS (assuming they port a dlm to linux).  Of
course, with a web server on a cluster, I would imagine you
would want dynamic DNS setup for the whole cluster all the
time, so if a node fell out during operations, the load,
which was split three ways, is still split two ways evenly. 
I am not sure what kind of special handy work is required,
if any, in wrt Dynamic dns though.  I guess it would be more
useful to be able to take a node out of the round robin
queue on a node failure (and, inversely, put it back in on
node reintegration).  

mtr



More information about the Linux-HA mailing list