Address Sorting NOT in V8

Jim Reid jim at rfc1035.com
Sat Apr 1 08:11:21 UTC 2000


>>>>> "Kevin" == Kevin Darcy <kcd at daimlerchrysler.com> writes:

    >>  Because relying on the order that the resource records in an
    >> answer from a name server being in a specific order was
    >> generally not a particularly smart thing to do. Think
    >> round-robin DNS for instance. Or resolvers that "sort" the
    >> addresses in an answer. Or resolvers that get one order of the
    >> RRset when they query one name server for some name but a
    >> different order if they query another name server for the
    >> self-same name. Or the name server forwards a query to some
    >> other name server that returns an answer that's optimised for
    >> the server that forwarded the query, not the original host that
    >> made it, etc, etc.

    Kevin> Jim, it works just fine on an intranet where all of the
    Kevin> nameservers can be configured with the same sortlists.

True, but intranets are not quite the same as the Internet because
things like the network topology and IP address allocation should be
more structured in the former. And even on an intranet I'm not
convinced that the name server should be sorting its answers. I think
that should be done by the resolver where whatever sort criteria are
considered appropriate - alphabetical, network routing, phase of the
moon - for the application that made the lookup. Why impose a "one size
fits all" approach from the name server? Sometimes this can be OK, but
not in the general case IMHO.

    Kevin> As for resolvers being configured with "wrong" sortlists which
    Kevin> interfere with the server sorting, how many users do you
    Kevin> really think are going to be configuring sortlists in their
    Kevin> resolvers anyway?

Who knows? It's amazing what some people do when they (mis)configure
their name servers and resolvers.

I think you miss my point: what if the name server's sorting of
addresses is wrong for some user or application? A global sortlist
might not be appropriate or even correct.



More information about the bind-users mailing list