On 7/5/2011 10:31 AM, John Levine wrote:
For an application that is likely to encounter a different IP address for
essentially every query, across a very large number of queries, the only
solution I see available is to use a different cache.
Seems reasonable. I gather that there are already caches with the
ability to partition themselves so that records from different
subtrees compete for different pools of cache entries.
I'm also sort of surprised that we don't seem to have all that much
experimental data about cache behavior. The MIT papers that Tony
cited are interesting, but they're also ten years old.
I doubt that partitioning according to DNS structure will work for this, since
the DNS has no formal semantics to different parts. Hard-coding knowledge of
particular tree segments can work for specific examples, in small scale, but
won't work for the constantly-changing IP Address behavior cited in this thread.
I believe that the solution is to have the applications, themselves, distinguish
the cache they are using (or the containing library). A blocklist app needs to
use a different library/cache than a web browser.
Ietf mailing list