Brian E Carpenter <[email protected]> wrote: >> Just watching the recording... >> >> Like Toerless, I am not sure what you are proposing to do exactly. >> How does what we need to do differ from what MIF (failed?) to do?
> Well, MIF did produce PVDs, but IMHO that was a completely different
> problem space. I think the correct question is how does the problem
> differ from the one solved by BRSKI+ACP, plus the missing bit of
> Anima, which is how to identify roles within the ANI and where the
> edge of the autonomic domain is. Our feeling is that this is a more
> general concept than just autonomics.
> (For example: Segment routing assumes that we know where the SR
> domain edge is. DETNET assumes that we know where the domain edge
> is. Diffserv assumes the same thing. Up to know, we only know
> because we configure the interior and edge nodes explicitly.)
I think that I comprehend the problem you described.
I don't understand the scope of the solution you imagine.
Is it terminology? Is it a new protocol?
Your point about how does it differ from the ACP boundaries is an interesting
one.
Let's consider a mature ISP or Data Center operator or Enterprise with a
fully functional ACP.
The DC operator has ToR switches that they manage, and their customers have
"things" in the cabinets that see a default route and stacks of IP-level
transit. Are the customer devices in the cabinets, or the desktop PCs
within the domain? Probably not for the customer devices... maybe for the
desktop PCs. Maybe not.
I think that for the set of scenarios your document outlines the answers vary
widely as well... quite possibly different scenarios could overlap, and yet
have different boundaries.
>> I was hoping that the ID would tell me more, but it didn't manage that
yet.
> OK, we'll try harder ;-). I will be giving these ideas a run at greater
> length at the APAN research workshop in Auckland on August 6th.
okay, that sounds like a good way to get feedback.
>> (When you say, which "node" is a member of the domain, I think you mean
to
>> say, "which interface on a node" is a member of the domain... ?)
> I think that depends on the scenario. (Do you think that
> individual interfaces should have an IDevID?)
Sometimes yes. Mostly, no.
Should they have different LDevIDs might be a more interesting question, and
I think having different LDevIDs might happen with some frequency in some
scenarios. Network Function Virtualization brings another layer to this.
(I don't expect BRSKI to bootstrap the other LDevIDs, but rather the ACP for
the physical devices would provide a mechanism by which virtual routers would
be created for customers)
--
] Never tell me the odds! | ipv6 mesh networks [
] Michael Richardson, Sandelman Software Works | network architect [
] [email protected] http://www.sandelman.ca/ | ruby on rails [
signature.asc
Description: PGP signature
_______________________________________________ Anima mailing list [email protected] https://www.ietf.org/mailman/listinfo/anima
