forwarders

[Kevin Darcy]

Bill Manning wrote:

> % * dns load balancing with multiple selectable algorithms
>
>         RR is predominate, others have been proposed and even
>         implemented. Daisy Chen did one for IBM that never
>         escaped, David Barr has had one in active use that
>         predates RoundRobin.

I think it might be useful to distinguish here between merely implementing a server-configuration-based selection of
sorting algorithms, versus actually coming up with a new record type which would allow the master's sorting preferences
-- whatever they may happen to be -- to propagate through the normal caching/referral channels. Did Chen or Barr
implement the former, the latter, or both?

> % * dns redundancy (like multimaster dns, or maybe primary dns failover, maybe like dhcp failover, but for
> dns).        See T.P.Briscos RFC on this.

That RFC would appear to be far more relevant to the load-balancing concern than redundancy of master servers.

Frankly, I don't see what could be so difficult about reconfiguring a slave to be the dynamic update master if the
"real" master should fail. It would be a little messier to "fall forward" once the real master was back up, but again,
is this an insurmountable scripting/programming problem? I think not.

As for how the clients know to send their updates to the "backup" master when the "primary" master is down, a
multi-address name for the master given out in "fixed" order should do the trick .Which gets us back into
RRset-ordering issues again...

Bear in mind, however, that I have very little operational experience with Dynamic Update, so take the comments above
with a grain of salt.

> % * Controlled subnet-sorting.  What I think is meant, is an
> %     advanced form of subnet-sorting, based on some if/then logic.
>
>         I have no idea what this means.

I think I understand it. But what would be required beyond what BIND's "sortlist" gives today?

> % * dns company merger facilities.  For example for specific domains or domains covered with regular expressions, or
> %     wildcards, select a specific dns server or servers, bypassing normal dns resolution behaviour.
>
>         See the DNAME work that is undergoing standardization. Available
>         in the BINDv9 beta trials now

Hmmm... DNAMEs, as I understand them, allow "aliasing" of a whole domain, e.g. every query you get for, say,
*.chrysler.com should be considered an alias for *.daimlerchrysler.com, where "*" is a "super-wildcard" that can go any
number of domain levels deep. But DNAMEs don't permit selective forwarding based on RE matches, do they? That's what
I think was being requested. Personally, I think selective forwarding based on RE matches could lead to very twisted
configurations and I'm not sure what benefit is hoped to be gained that is not already available through the per-zone
forwarding and per-domain "de-forwarding" already provided by BIND today.


- Kevin