Re: [DNA] Comments on draft-shim-dna-proactive-00.txt

[Bernard Aboba <aboba@internaut.com>]

> That may mean a single pair of messages direct to that router, but
> I'd guess that this is fairly fast if the configuration was correct
> (a likely case).

Exactly.  One thing to keep in mind is that a hint can both be highly
likely to be correct, and also not worth considering, depending on the
penalty paid for its being wrong.

As an example, NS/ND (or ARP) can only demonstrate reachability, not
unreachability.  Therefore a hint indicating that the host has remained on
the same subnet will have a substantial downside (a timeout) if it is
wrong.

As an example, in DNAv4, assume a DHCPREQUEST would require DHCPR_TIME to
determine if a host has remained on the same subnet, while a
reachability test would require REACH_TIME.  If we wait
ARP_TIMEOUT before sending a DHCPDISCOVER, which requires DHCPD_TIME to
complete, then if a hint that the host has remained on the same subnet is
wrong x fraction of the time, then it is only worth considering at all if:

DHCPR_TIME = (1 - x) * REACH_TIME + x * (ARP_TIMEOUT + DHCPD_TIME)

x =          DHCPR_TIME - REACH_TIME
      -------------------------------------
      ARP_TIMEOUT + DHCPD_TIME - REACH_TIME

Let's assume that DHCPR_TIME = 100 ms, REACH_TIME = 10 ms, ARP_TIMEOUT = 1
sec, DHCPD_TIME = 200ms.  Then

x = (100 - 10)/(1000 + 200 - 10) = 90/(1190) = 7.5%

So the hint only needs to be wrong 7.5% of the time before it is no longer
useful.