random question about the internals of bind

[Barry Margolin]

In article <86nlar$hv$1 at net.indra.com>, Kirk Johnson <tuna at indra.com> wrote:
>i'm curious if anybody can tell me what happens to this state (the
>"approximate" round-trip times associated with each NS record) when
>the TTL on the NS records expires. when the same group of NS records
>are refetched, is any of the state that was associated with the
>previous instance of those NS records (before the TTL expired)
>preserved? (this is the simple case -- assume that bind gets back
>exactly the same NS records as it had previously.)

I believe the state is associated with the server's IP address, not the NS
record.  So the expiration of the NS record doesn't have any impact on it
-- there could be other NS records for other domains that point to the same
server.

The presumption in this design is that response time is mainly affected by
server load and network topology, not anything zone-specific.  This isn't
true in the case of some lame delegations (they may cause very long
response times, due to the server having to perform a recursive query of
its own), but I think lame response times may not be included in the state.

>bonus question: what happens in the same case if, upon refetching the
>NS records for the zone in question, those NS records have changed
>(where there may be _some_ overlap with the previous set, but perhaps
>having dropped a few records and added a few others)?

Since the state is associated with the server's IP address, not the NS
record for a specific zone, this is pretty simple.  When it wants to look
up something in that zone, it will find the response time state for the
servers in the current NS record set, and pick one of the ones with low
response time.

-- 
Barry Margolin, barmar at bbnplanet.com
GTE Internetworking, Powered by BBN, Burlington, MA
*** DON'T SEND TECHNICAL QUESTIONS DIRECTLY TO ME, post them to newsgroups.
Please DON'T copy followups to me -- I'll assume it wasn't posted to the group.