> As per thinking that the external root-servers are EPOF, well
> yes, but 13
> going AWOL at the same time? hmmm.... If they DO go awol at
> the same time
> for a given ISP, it's basically because that ISP has lost all
> his external
> connectivity, which makes it useless anyway...
I have 6 documented cases of exactly this event. The entire
root-servers.net sytem crashed six times in 1998. All six were
documented in NANOG discussions. The outages were "days" in duration, in
each case. Just recently (within the past two weeks) we saw a case of
cache contamination that took out all 13 root-servers, save 2. Ergo,
your statement is false. These root-zone outages occur with more
frequency than you assume. When they do, no site that is not using their
local root-zone, remains operational, even for internal stuff. Every
machine that needs to go to the root-zone will hang. The NANOG
discussion definitely centered around local root-zone copies (daily
retrieval) for all the major backbone providers (NANOG) as a "best
practice". You should start lurking on NANOG. The issue of alternate
TLDs is irrelevant in this discussion. The side benefit is that this
reduces the load, from each ISP, on the root-servers.net system, by the
number of clients on each ISP.
> Statistically, I think you'll find your affirmation of "every
> major ISP does
> exactly this [maintain local copies of the root-zone]" false.
> At least info on hits that the root-servers get (once deducted the
hits for
> SLDs) should prove this so.
Considering the traffic I saw, on NANOG, regarding exactly this issue,
during the entirety of 1998, I believe that I am reporting accurately.
Sean Donnelson has been making this recommendation and he is the outage
guru.
Every portal-play and ISP, that I design system architecture for,
eliminates name server EPOFs exactly this way, whether they also access
alternate TLDS (some do, some don't) is irrelevant.