[dnssec-deployment] cache performance

[bmanning at vacation.karoshi.com]

> On Fri, Jun 27, 2008 at 02:10:03PM +0000, Paul Vixie wrote:
> > From: Scott Rose <scottr at nist.gov>
> > 
> > 1.  Caches will grow even larger (if resolvers query for names in these new
> > TLDs)
> 
> cache size depends
> on overall namespace size, and then as you say, working set size.  neither
> the overall namespace nor the working set will be made different just because
> the shape of the namespace (more TLDs) changes.

	caches do seem to follow zipfian curves, from studies done
	over the past 15 years.  so there is a correlation btwn 
	the size of the namespace and the size of any given cache.

> i'm on record, many times over the last decade, that a million TLDs is not
> going to change the root server system's provisioning, nor the average joe's
> cache size, at all.  it's only when names get used that any of this matters,
> and most of the traffic hitting the root name servers is complete garbage and
> is probably going to continue to be complete garbage.

	rss provisioning == I agree.
	average joes cache size == perhaps not, but the data is not in.
		just because the roots get traffic does not reflect directly
		on AJC's cache behaviour.

> (the fix to that would be if nobody could be on the internet if they didn't
> properly implement negative caching, and, if NXDOMAIN responses could say not
> just that the name you queried doesn't exist, but that the TLD you're trying
> doesn't exist, so please negatively cache that.  neither of these things is
> ever going to happen, so, the root servers are massively overprovisioned, and
> within the headroom we have to maintain to handle the garbage, a million more
> TLDs would be noise by comparison.)

	assuming facts not in evidence.  there are also effects of relative bandwidth
	betwn and authoritative server and the cache.  

--bill