Primary DNS failover

[Kevin Darcy]

I've heard that Microsoft's DNS implementation allows dynamic updates to
be received for a given zone by multiple servers, hence "multimaster".
Then the changes are replicated between masters, using presumably some
sort of convention (hopefully configurable) for dealing with the
inevitable replication conflicts.

I think that with a little scripting and/or other enhancements, BIND could
theoretically be made to act in a somewhat similar fashion: under normal
circumstances a slave will forward dynamic updates to the master
transparently to the clients; if the master fails, the slave could detect
this, reconfigure itself as the master and accept updates itself; then
when the master comes back up, it would initially come back as a slave to
the zone, until it was up to date on all of the changes, at which time the
master/slave roles would reverse and things would be back to normal. Any
volunteers to write this?

Note that "multimaster" DNS, as Microsoft implements it, or along the
lines I just described for a BIND implementation, would not require any
changes to the base DNS protocol itself. It would, however, be nice if
there was an open standard so that different products could all
interoperate in the same multimaster framework.


- Kevin

Joseph S D Yao wrote:

> On Thu, Mar 02, 2000 at 05:18:34PM -0800, Michael Voight wrote:
> > Microsoft uses multimaster. (I think that's the name)
> > Obviously, not being standards based, you can't mix it with a UNIX
> > primary for the same domain.
>
> Over the last month, this has been mentioned numerous times by two
> people, with no explanation of what it is.  Michael, what is this
> "multimaster" capability?  What does it provide over and above the BIND
> capability of multiple equivalent name servers that are "failed over"
> to automatically when any are down?  If it distributes the database of
> a SINGLE domain over multiple servers, how does it avoid requiring
> searching ALL servers before declaring that a given name does or does
> not exist, and all the other problems that have been mentioned before?
>
> I do have to admit that Microsoft's desire to extend a standard is
> occasionally for the purpose of giving a legitimate capability that is
> not otherwise available.  But often it is solely to lock in the user -
> I suspect they would have embraced EBCDIC if they had thought they
> could get away with it.  ;->
>
> Thanks.
>
> --
> Joe Yao                         jsdy at cospo.osis.gov - Joseph S. D. Yao
> COSPO/OSIS Computer Support                                     EMT-B
> -----------------------------------------------------------------------
> This message is not an official statement of COSPO policies.