Re: Alternative to BGP-4 for multihoming?

[Patrick Evans <pre () pre org>]

On Tue, 14 Mar 2000, Pete Templin wrote:

> On Tue, 14 Mar 2000, Marc Slemko wrote:
>
> > Saying "the clients's primary DNS" is misleading.  There is no way to know
> > what the "primary" DNS server is for a zone, and there may not even be
> > what is typically known as a primary.
>
> What if it's a UUNet resold modem to a client of iAmerica - what server
> gets used then?  We know that UUNet's DNS servers are likely to not be
> located close (in net terms) to the client, and how do we know what DNS
> servers are being assigned to the client? 
User's machine contacts caching nameserver x to do a lookup.
Nameserver x contacts authoritative nameserver y, which then works out
where x is before returning an RR that's good for wherever x is. 

If n isn't net.near to the user's machine, then something's a bit
weird. If part of a dialup ISP's internal network falls over, you
hardly want every single user's resolvers to fail!

> Or what if my clients get assigned dns servers in 192.168.254/24?  Sounds
> to me like it's not a valid geographic identifier.
I'd hope that nameserver would talk to the world with a real source
address, which y would then use to do a proximity test, rather than a
1918 address - at least it should if it actually wants to get a
response. What address the client-side interface uses is neither here
nor there.

-- 
Patrick Evans - Sysadmin, bran addict and couch potato
pre at pre dot org                     www.pre.org/pre