[rrg] NTP and first packet delivery

[William Herrin]

On Sun, Nov 30, 2008 at 11:52 PM, Scott Brim <swb at employees.org> wrote:
> The issue will rarely come up.  In order for it to arise, a site
> border router must come up cold with no knowledge of prefixes where it
> wants to find its NTP servers.  Once it maps these prefixes, which
> will very probably be few and well-used, it will have mappings for
> them until it comes up cold again.  So as yet another alternatives,
> avoid completely cold restarts, or give it what it needs via
> configuration or synchronization.  There are probably other
> possibilities.

Scott,

It need only expire the cached map entry for inactivity. Which it will
tend to do over tens of minutes with no use and plenty of competition
from other maps, something typical of the traffic pattern for NTP. Nor
can you store the cached map for long: in a pull-cache system, cache
expiration is your only guarantor of convergence. Everything else you
throw at it is a heuristic that only works *most* of the time.

That's one of the things that makes NTP good for illustrating this problem.

I remind you about:
http://www.ops.ietf.org/lists/rrg/2008/msg00465.html

and the thread at:
http://www.ops.ietf.org/lists/rrg/2008/msg00509.html

Also, look at: http://bill.herrin.us/network/statechange.html for an
explanation of why you can't cache entries for very long.

Regards,
Bill Herrin


-- 
William D. Herrin ................ herrin at dirtside.com  bill at herrin.us
3005 Crane Dr. ...................... Web: <http://bill.herrin.us/>
Falls Church, VA 22042-3004