Fail Over name serving using BIND

[Len Conrad]

>1) How exactly are the ROOT servers deciding which NS to give to the client,
>I have heard 3 different things, and can not find anything on ICANN,
>Internic, RFC 1035, or the email archives of these ISC lists that spell this
>ROOT server behavior.
>     In,
>         NS    ns1.foo.com
>         NS    ns2.foo.com
>         NS    ns3.foo.com
>     a) does the ROOT server poll all 3 and give the fastest response to the
>client's query?

The root servers don't know whether the machines of the NS and A records 
they pass out are responding or not, and certainly don't worry about the 
speed of those that are alive.

>     b) does the ROOT server poll them in line order handing the first
>response to the client's query?

the "client" is another, querying DNS that does all the active, iterative 
querying/navigating the namespace to obtain an answer, if it exists.

>     c) does the ROOT server "cycle" through the entries using them all
>equally?

yes, but as soon as the root-servers' answers are cached in another DNS's 
cache, it's the caching DNS that decides the physical ordering of those 
same records in its own response packets.

>2) In BIND, is there a way to assign a weight (like the MX weight) to each
>NS entry, so that one will be used before the others, unless it is
>unreachable, then the ROOT server pulls the next lowest weight?

no. The TLD and root-servers do not query, they only give out postive, 
negative answers and referrals.

>any information, hints, links, or constructive profanity would be greatly
>appreciated,

all the functions you erroneously think are done by remote DNS's are 
actually done by the DNS that has a query to resolve.

Get the Oreilly DNS + BIND book, 4th edition, or find on cdrom on Oreilly's 
Networking.

Len