SEAN at SDG.DRA.COM
Sun Aug 18 19:20:27 UTC 1996
>Of course, we'll be burning much more address space than we 'need' to,
>but we'll be adding 0 routing entries, even though we are multiply
>multi-homed. However, as I'm only a registry person and I "don't run
>real routers", it is possible I am missing something obvious that will
>result in this not working. If so, please let me know before I turn
>APNIC into slag...
What's the scarce resource we're trying to conserve this week?
There is a provider in St. Louis which is trying to do something similar.
Since people are buying it, the theory must sound good.
In practice it makes things worse because you turn DNS into the
router table. As DNS round-robins through the different provider
addresses, you'll get addresses with the source and destination
address pair from the same (or shortest path) provider to
addresses pairs which need to transit another (or longest path)
provider irregardless of the actual state of the network. Depending
on the transit agreements between the providers to "heal" discontinuities
doesn't work well.
What's even worse, you have to depend on the quality of programmers.
Since many programs only try the first IP address in the DNS response,
you'll get more failures as the DNS round-robins through a down provider's
address block. Unless you start make dynamic changes of your DNS records
based on your routing state, blech. Maybe I missed something, and Paul
Vixie intends to integrate BGP-5 and Bind-5.1.0.
Under the heading, history repeats itself, sortlist isn't much of a
Sean Donelan, Data Research Associates, Inc, St. Louis, MO
Affiliation given for identification not representation
More information about the NANOG