on 7/31/2002 12:07 PM Vernon Schryver wrote:
>>From: "Eric A. Hall" <ehall@ehsco.com>
>>...
>>As to your point, I will simply reiterate that the function remains to be
>>bound by frequency, and thus the quantity of queries over time. The number
>>of cache misses will not double unless and until you double the number of
>>out-of-window queries. A TLD which never gets queried has no misses.
> That particular reiteration of Mr. Hall's thesis contains two valid
> statements that are either independent of his original thesis or that
> contradict it.
Mr. Hall's original and current thesis is that adding TLDs as an act is
harmless, and that the real point of concern is the number of queries.
This point has not yet been disproven. The counter-argument that a busy
("popular") TLD will result in more queries only proves the point,
although the advocates of such an argument refuse to admit that the
frequency of queries -- the number of queries over time, the measure of
"popularity" -- is the issue, not the number of TLDs.
> As far as I can tell, Mr. Hall supports the notion that dividing one
> of the current large TLDs into N smaller, approximately equal sized,
> each quite large TLDs would not cause an approximately linear increase
> of N times as many hits on the roots.
Mr. Hall believes that it would be N-minus, since not all networks will
query for all of the subsequent TLDs, or will do so at a lower rate (such
as five days instead of two days). The resulting load would be more than
the original consolidated TLD (never argued), but it does not follow that
it would be a multiple of the original load.
> An equivalent statement is that
> combining N of the current large TLDs into a single TLD would not
> reduce the number of queries for the roots by a factor of N.
That is correct, with the right qualifications. A consolidated TLD (such
as grouping all ccTLDs into a .nation TLD) would result in fewer queries
to the root, but the corresponding increase in queries for that TLD by the
end-systems would also mean that cache misses were discovered sooner
(every 48 hours exactly).
> Mr. Hall's thesis has been disproven in several recent comments.
Mr. Shryver is easily impressed by the use of "Wrong".
> think one of those comments came from Mr. Hall himself. That comment
> was when he noted that his server's cache had entries for both .com
> and .edu. Those two entries required his server to query the roots
> twice. If .com and .edu were merged, then Mr. Hall's own servers
> would have made half as many queries of the roots.
But Mr. Hall's cache would also make that one query more frequently.
--
Eric A. Hall http://www.ehsco.com/
Internet Core Protocols http://www.oreilly.com/catalog/coreprot/