Re: Anycast root metrics and analysis

[hardie@equinix.com]

kre quotes me, then writes
>   | If the current plan resulted in worse performance to
>   | end sites, I believe it to have failed.
> 
> Absolutely.   But there is a difference between "worse performance
> than now" and "worse performance than could theoretically be obtained".
> The former is what we need to guard against, not the latter.

We are examining a change to the operation of a pretty fundamental
part of the Internet infrastructure.  I agree that we need to guard
against "worse performance than now", but I also believe that we need
to ensure that the fix we employ is one which gives the best bang
(performance) for the buck (operational pain).  It is mighty unlikely
that we will get apply more than one fix to this particular problem.
To me, that means a test of how well we can get this system to perform
is important.  There is also the tricky question of performance for
whom, and a fair likelihood that the results of that will not be an
even distribution.

In my first email, I proposed a change to Masatka's proposal which
would allow as many local copies of a specific anycast root as were
desired, but restricted transit to them to a single well-connected
provider (as is the current case).  I have heard you say that you
don't believe that the problem I identified is important enough to
warrant further testing and that you believe that the increased
distribution of anycast root server instances would solve the problem
anyway.  Without further data, I doubt we will agree on the first
claim.  Any chance that you can run an aroot instance from where you
are so that we can gather more data to prove your second point?

				regards,
					Ted