RE: 6MAN WG Last Call:draft-ietf-6man-ipv6-subnet-model-00.txt

The host has 4001:420:3800:809::/64 as a prefix in host PrefixList
because the RA sent this prefix as on-link. The host has cached on-link
determination for this prefix and the current lifetime of the prefix is
much longer than the time it takes the host to reboot. Now the host is
rebooted and during this reboot, the admin of the network renumbered the
prefix to expire and also configured a new RA where RA has no PIO. After
reboot, the host still continues to use this cached information because
the prefix hasn't timed out and host did not see the renumbering RA.
After reboot, when the host sees a new RA the host is supposed to update
information from. RFC4861 says to use union of information from
different RA's, except where MTU or Lifetime of a prefix is concerned,
to take the new information only. But the new RA has no PIO and hence no
Lifetime. So it gets murky. Do we keep a union of 4001:420:3800:809::/64
as on-link and all other non-link-local destinations as off-link due to
an RA with no PIO or do we consider the new RA with no PIO to signal
Lifetime for any non-link-local prefix as zero?

What if the cache gets corrupted and the corruption changes a few bits
in the cached prefix and even a reboot doesn't fix the changed bits. The
RA remains same between reboot. But a totally different prefix now
exists as on-link. Why is this not a problem?

Also interesting is the fact that Suresh Krishnan who is author of DNA
protocol and Chairs DNA blessed our bullet 3 on June 18th, 2008 to
prohibit caching of on-link determination but Thomas is reporting that
DNA proposes the caching we disallow. 

Hemant

-----Original Message-----
From: Thomas Narten [mailto:narten@us.ibm.com] 
Sent: Wednesday, July 09, 2008 9:30 PM
To: JINMEI Tatuya / ????
Cc: Wes Beebee (wbeebee); Hemant Singh (shemant); Brian Haberman; Bob
Hinden; ipv6@ietf.org
Subject: Re: 6MAN WG Last Call:draft-ietf-6man-ipv6-subnet-model-00.txt

Note: replying to a number of separate messages here...

JINMEI Tatuya / =?ISO-2022-JP?B?GyRCP0BMQEMjOkgbKEI=?=
 <Jinmei_Tatuya@isc.org> writes:

> At Thu, 3 Jul 2008 17:07:34 -0400,
> "Wes Beebee (wbeebee)" <wbeebee@cisco.com> wrote:

> > <wb>
> > What if there are cache-inconsistency problems, prefix renumbering, 
> > or stale information?  I think it's better just to get rid of 
> > caching on-link information in stable storage and get such 
> > information fresh from RA's.  That way, administrators can better 
> > rationalize the behavior of their network from the configured RA's.
> > </wb>

> The same argument applies to caching the address, so it cannot be a 
> reason why we specifically prohibit the (on-link) prefix part of this 
> behavior.

Exactly.

> Actually, I see it can rather be harmful if we only prohibit the 
> on-link prefix part of this behavior.

Again, I agree.

Note also, that DNA approaches essentially "cache" information about the
configuration of an interface across interfaces going up and down
-- which is very similar to the type of caching you are proposing to
disallow.

There is nothing wrong with caching information (as long as it has not
expired) so long as one replaces it if updated information comes along
later.

So I continue to be unpersuaded by any of the discussion so far that we
should saying "don't do that". 

Thomas

--------------------------------------------------------------------
IETF IPv6 working group mailing list
ipv6@ietf.org
Administrative Requests: https://www.ietf.org/mailman/listinfo/ipv6
--------------------------------------------------------------------