DNS question from a network guy
Mark_Andrews at isc.org
Mark_Andrews at isc.org
Wed Jan 22 00:55:02 UTC 2003
> <Mark_Andrews at isc.org> wrote in message news:b0kc1p$2ecg$1 at isrv4.isc.org...
> >
> > > I am not a DNS guru. I did search the archives first...
> > >
> > > If you have two or more records for a host, DNS will use round robin in
> its
> > > replies... What is the behavior if you delegate a domain, for instance
> > > www.mydomain.com to two name servers?
> > >
> > > in the mydomain.com name server...
> > >
> > >
> > > www in ns ns3.mydomain.com
> ...
> > Well they will be handed out in the order specified in named.conf.
> > The caching servers will use which ever nameserver that responds
> > fastest. The rtt estimates will be computed as a side effect of
> > normal query traffic.
> >
> > Mark
>
> Don't all NSs get a random RTT below anything that would occur in normal
> operation. This way, all NSs will eventually be used and have an actual RTT
> calculated. The lowest RTT will be used first. Is this still correct? If
> so, I assume round-robin or subnet sorting would have no effect on NS
> selection? tia
BIND 8 uses bands 64 ms wide with the first band being
32 ms wide, topology is a second order effect.
BIND 9 just uses the rtt estimate.
RTT estimates of unused servers drop on each query which
eventually causes named to try them.
RTT estimates of server that fail to respond are raised
reducing the probability of the server being used first for
the next query.
RTT estimates of the current server is adjusted based on
current RTT and previous RTTs.
Unless you are measuring at cpu rates there are no RTT
estimates that cannot occur including zero. Sub milli-
second RTT are possible as are sub microsecond RTTs. RTT
is normally measures in milliseconds as that best fits speed
of light around the world. It can also cope with multi second
RTT responses that occur with some nameservers.
Mark
> -- wjs
--
Mark Andrews, Internet Software Consortium
1 Seymour St., Dundas Valley, NSW 2117, Australia
PHONE: +61 2 9871 4742 INTERNET: Mark.Andrews at isc.org
More information about the bind-users
mailing list