And what happens to ".re" relative to ".com" has nothing to do
with the story.
.edu versus .com then, or .au versus .com, or anything you
want. There are
millions of organizations that don't issue queries for those
zones either.
I am pretty sure that the load caused by ".org" to the root is almost
equal to the load of ".com", and that ".edu" is pretty close. I would
suspect that .re is . According to www.netsizer.com, there are about
200M entries total in the DNS. The most popular domains are .Net (67M),
.Com (42M), .Edu (8.7M), .JP (8.5M).
A way to look at the root impact issue is to assume a resolver that
starts with an empty cache, and starts putting out queries at random. We
will assume that the queries' targets are picked at random from the
entire population of hosts; this is not entirely correct, but this is
the only assumption we can make with the data at hand. Suppose that the
resolver's client make N queries during a standard "time-to-live"
interval. The probability to hit the root for a given domain is
proportional to the probability that at least one of the N queries hit
the specific domain, which is a function of N and of the fraction X of
hosts that belong to this particular domain: P(1 hit after N queries)=1
- (1-X)^N
I ran the number. They are pretty interesting. I picked here 5
significant domains, and computed the chances that a given resolver asks
them once to the root after passing 100 queries (N=100), 1000 queries,
10000 queries:
Domain Hosts 100 1000 10000
.com 42M 100.0% 100.0% 100.0%
.jp 8.5M 98.7% 100.0% 100.0%
.fi 1M 40.2% 99.4% 100.0%
.cn 124K 6.0% 46.2% 99.8%
.su 22K 1.1% 10.6% 67.2%
In short, there is not a lot of difference between the very large domain
(.com, .net) and the moderately large (.edu, .jp); even medium size
domains such as .fi generate almost as much load if you assume a
resolver will issue 1000 queries before the cache expires; with this
hypothesis, the load generated by a rather small domain like .cn (there
are 40 larger domains) is almnost 1/2 of that of .com! Indeed, if you
assume that the caches will be good for 10,000 queries, even the smaller
domains, like the obsolete .su, will cause a significant load.
-- Christian Huitema