Just blending the two response sub-threads:
On 12/07/2016 00:37, Laurent Ciavaglia wrote:
> Hello,
>
>> Yes. There's been an assumption, I think, that one "autonomic function" ==
>> one ASA.
>> We need to be clear if that is an axiom, and we need to think about how ASAs
>> are
>> named, and if those names need to be registered somehow.
>
> Mmmh... where does this assumption come from...?
> I think we've been quite clear the general case is one AF == multiple ASAs.
> An AF instantiated by (only) one ASA is a sub-case.
> Cf. figure 1 of the Reference model draft.
>> Yes, that misunderstanding keeps popping up all the time. I think RFC7575
>> is quite clear:
>>
>> Autonomic Function: A feature or function that requires no
>> configuration and can derive all required information through self-
>> knowledge, discovery, or Intent.
>>
>> Autonomic Service Agent: An agent implemented on an autonomic node
>> that implements an autonomic function, either in part (in the case of
>> a distributed function) or whole.
>>
>> Example: There is the "autonomic function" "bootstrapping of new nodes". It
>> consists of 3 different ASAs: The new_device ASA, the proxy ASA and the
>> registrar ASA.
> Michael: in your example, the agents represent different parts of the AF
> functionality.
> There is an additional dimension: an AF can be deployed over multiple
> resources(/devices). An ASA, encompassing the whole AF functionality, would
> then be instantiated on each device. E.g. a load-balancing AF deployed over a
> set of routers (each router will get an ASA of the load balancing AF).
Right. So an AF can have multiple ASAs in one node, as well as multiple ASAs on
different nodes. And if a section of Intent is aimed at a particular AF, then
the AF needs a name. But the AF is *realised* by a set of ASAs, which must
somehow be associated with the AF's name (otherwise they could not select
the relevant Intent).
>
> Why would ASAs need a name? That ASA need an ID is ok, but machines don't
> care about having stuff referenced with names... and I
> don't think human operators will have to deal with ASAs (one of the goals of
> ANetworking), or at least not often enough to
> justify the use of a naming scheme
Well, I used the word 'name' in a computer science sense ;-). Indeed it could be
a binary number or it could be a string. I don't see any harm in making it a
string,
for the sake of debugging convenience (as I have in my prototype) but it isn't
required.
I remind you that we have so far ducked the question of authorization
of ASAs to manage particular objectives.
> AFs could have names (and would actually need more than just name, e.g.
> version number, description of what it does, requires,
> etc.).
>
> Or am I completely overlooking something?
No, I don't think so, and some of this needs to be captured for the Reference
Model.
Regards
Brian
>
> Best regards, Laurent.
>
_______________________________________________
Anima mailing list
Anima@ietf.org
https://www.ietf.org/mailman/listinfo/anima
