Re: [DNA] Prefix information for link identification in DNA

[Greg Daley <greg.daley@eng.monash.edu.au>]

Hi Bernard,

I've got some ideas which may assist moving forward.
They're listed below.

Brett Pentland wrote:
> Bernard Aboba wrote:
> 
>>> There are certainly adhoc networks in which there is no router.   
>>> However, detecting attachment to such a network is quite difficult, 
>>> because nodes may join and leave and therefore there is no L3 
>>> invariant.  That is why the DNAv4 reachability test cannot be used to 
>>> detect attachment to adhoc networks, but rather adhoc attachment is 
>>> concluded after failiure of all other approaches (reachability test, 
>>> DHCPv4, etc.)
>>
>>
>> James Carlson has recently pointed out that there really is no 
>> intrinsic reason why DNA cannot be applied to networks without a 
>> router, and has proposed modifications to DNAv4 to allow this.  James 
>> has proposed that DNA be able to use any peer that is known to be 
>> stable as an indicator of network attachment.  This could be a router, 
>> or it could be another device.  About the only fundamental limitation 
>> is that "network anchor" probably shouldn't obtain its address 
>> dynamically.
> 
> 
> OK.
> 
>>>> I would also agree that there are situations in which a network 
>>>> cannot be named using one of its prefixes.  In DNAv4, a private 
>>>> network is not suitable for identification because it is not unique.
>>
>>
>> Along these lines, James has also proposed that confirmation of a 
>> "network anchor" enable the plumbing of all network configurations 
>> relating to that anchor.  The result of this is that networks are not 
>> named using their prefixes;  they are named based on the IP 
>> address/MAC address of the network anchor.
> 
> 
> They might not need to be named that way either.  If there's some way
> to select the network anchor then that device might be able to select
> an identifier that is used for naming.  That way if the network anchor
> goes away and is replaced by another, the name needn't go away with it.
> 
>> It seems to me that James' comments apply equally to both DNAv4 and 
>> DNAv6 and call into question some of the fundamental assumptions that 
>> have been made relating to DNA.
> 
> 
> Decoupling DNA from both routers and prefixes has some attractions; 
> stability of naming and reliance on routers, for example.  I guess we
> need to look at the cost, and perhaps more importantly how a scheme
> based on these network anchors would interoperate with devices/links
> that don't have any support for such a scheme.


James' idea of decoupling the link identification from both routers
and prefixes does sound like an interesting idea, but I think there
are some practical constraints which affect the performance of
such a system on non adhoc networks.

I guess that my primary interest is identifying change for hosts
with ongoing sessions on the Internet.   This change can then be
used to undertake mobility operations and re-establish packet flow.
Where no router exists, the primary value upon change is in
stopping packet transmissions (as they won't succeed), and notifying
applications.

Changes to how link-change detection works in adhoc networks
with routers depends on the IPv6 subnet model being used though.

Previously we looked at unicast reachability detection for
ND hosts, using NS/NA to a device which was known to be on-link.
This allowed fast checking when the addressing and routing
configuration were still valid, but required timeout to identify
that link-change had occurred.   From what I can tell,
any scheme which uses unicast reachability probing requires
timeout before detecting if change has occurred.

Therefore, it's necessary to make use of multicast discovery
mechanisms in order to rapidly detect change.

For non adhoc hosts, Router Discovery makes a good transportation
mechanism for the link-identification information, even if the
identifier is not associated with a prefix.  Its multicast discovery
capabilities allow any identifier advertiser to immediately
identify change.

For adhoc hosts, it may be possible to use a non-prefix identifier
associated with the link.  This would presuppose adhoc hosts participate
in a routing protocol discovery which could pass such an identifier
though.

In order to supply authorization for the link-identity, this
identifier may be tied to the Internet gateway, with adhoc nodes
readvertising the link identifier sent in RAs from the router.
Authorization information could either be sent by adhoc nodes
or refer the host directly to the router.

Where no router exists, the identifier could be supplied by
a network anchor, although this may not be able to provide
explicit authorization.

If we assume that it's OK to continue to operate with IPv6
Router Discovery for non adhoc nodes, it still makes sense
to allow for introduction of non-prefix link identifiers for
these and other purposes.

To that end I suggest that DNA WG:

* Identifies if the IPv6 Subnet model for adhocs
   requires non-prefix link identifiers.

* Allow for non-prefix link identifiers to be used and tracked
   by DNAv6 hosts

* Define message formats which allow non-prefix link identifiers
   to be sent in RAs.

* Investigate requirements for configuring non-prefix link identifiers
   as a (separate) extension to the DNAv6 specification.

* Investigate requirements for passing link identification in adhoc
   networks.

The second and third points can be done initially without
clarification of the other points, so non-adhoc hosts
can still detect change using these identifiers, even if they
don't understand the special meaning of these identifiers
from a routing protocol perspective.

Does this make sense?

Greg