pre heat cache

Mark Andrews marka at
Thu Feb 18 19:55:58 UTC 2016

In message <20160218185957.GA2298 at>, Robert Edmonds writes:
> Tony Finch wrote:
> > Tony Finch <dot at> wrote:
> > >
> > > Funnily enough I recently wrote a tool to do this but I have been failing
> > > to publish a blog article about it... Have a look at this:
> > >
> > 
> > Longer blurb now published at:
> A large proportion of records are only ever used "once, or a handful of
> times", according to researchers:
>   ("DNS Noise: Measuring the Pervasiveness of Disposable Domains in
>   Modern DNS Traffic")
> If there were some way to only dump the most popular records from the
> cache (say the top 1,000 or 10,000), it would probably speed up the
> pre-heating process significantly and still warm the cache to an
> acceptable degree.
> Does rndc dumpdb dump the cache in DNS tree order, or could it be
> convinced to follow the LRU order?

Or realise that it isn't worth the effort to do this.  The DNS is
actually fast enough to not have to warm the cache.

If one wants to spend time fixing things go through the Alexa top
1M and complain to every operator that still has a server that
returns FORMERR to a EDNS query.  That will have more benefit than
any cache warming code and nameservers have to constantly re-learn
that a server doesn't support EDNS and that takes multiple round

Go through the top 1M and complain to those that don't respond to
unknown QTYPE e.g. TYPE666 or respond with NOTIMP/REFUSED or respond
with a invalid negative answer (e.g. wrong SOA, NXDOMAIN instead
of NODATA).  Getting rid of broken servers will provide much more
benefit then saving a couple of seconds when a server is restarted.
Not only do these errors cause slow lookups, they have put back
application development years.


> -- 
> Robert Edmonds
> _______________________________________________
> Please visit to unsubscribe
>  from this list
> bind-users mailing list
> bind-users at
Mark Andrews, ISC
1 Seymour St., Dundas Valley, NSW 2117, Australia
PHONE: +61 2 9871 4742                 INTERNET: marka at

More information about the bind-users mailing list