Hi Andy,

Thanks very much for the feedback.

In my view, I2RS is between the I2RS Agents and I2RS Clients.  An I2RS
Agent is associated with a single Routing element.
The I2RS agent can communicate with many I2RS Clients; an I2RS client can
talk to many I2RS agents.   I don't think that an
I2RS agent talks to other I2RS agents.

A routing element might have an I2RS agent and an I2RS client associated
with it.  For instance, say there's a routing element
that implements BGP and doesn't have a forwarding plane associated with it.
  The associated I2RS agent might support the BGP
service.  The associated I2RS client might be used to communicate to a
different routing element for its RIB service.

If a network application needs routing state at multiple routers, then I
agree with Adrian.  Either the i2RS client talks to each router
individually or the I2RS client talks to a single router to trigger the
distribution of the routing state (i.e. an RSVP-TE LSP, a prefix in OSPF,
etc).

I do hear you on the thoughts about a base core of functionalit y.  What
I'm picturing is that there are basically two types of info-models that
we need to create.  The first type are "per service" info models - where it
applies to a particular routing functionality - whether BGP, OSPF, ISIS,
RIB, PIM, RSVP-TE, etc.   We could use a better term than the overloaded
service - any suggestions?

The second type will be I2RS protocol-related info-models.  This might be
to describe I2RS-related event notifications or to control the role of a
client or to describe the capabilities of the I2RS agent in terms of what
services it supplies.  We haven't really touched on this part yet...

Each routing element may provide a different set of services - and I don't
think we can mandate what those are.  It'll depend on the role of the
routing element in the network.  What I think we can require is a set of
core I2RS protocol-related functionality - so that I2RS clients can learn
what is supported, the associated capabilities, etc.

Focusing on a few services at first, from a WG perspective, does make sense
to me.  I believe that is starting with the RIB information model and the
topology information model.  I think we need a simple BGP info model soon
as well - though I'm not sure if that's being worked on yet.

Regards,
Alia




On Sun, Jun 23, 2013 at 5:11 PM, Andy Bierman <[email protected]> wrote:

>
>
> On Sun, Jun 23, 2013 at 1:35 PM, Adrian Farrel <[email protected]>wrote:
>
>> Andy,
>>
>> > my question is how many I2Rs agents does a client need to
>> > contact to perform a task that requires changes on multiple routers?
>>
>> You probably won't like my answer "that depends".
>>
>> If the effect of the action on router 1 is to cause router 1 to use a
>> routing/signaling protocol to make things happen in the network, then the
>> answer
>> may be 1.
>>
>> If the changes cannot be effected in that way, then the client may need
>> to talk
>> to multiple routers to make the different things happen on each router.
>>
>>
>
> I meant a high-level task performed by the client that requires
> routing state to be manually injected on multiple routers,
> not via routing protocols.
>
> I am happy to consider the I2RS agent to be running on a single router
> and the I2RS client to be the network-level broker.  The API to
> specific network-wide services topology or routing can be standardized
> next.
>
> IMO it is better to start with the core functionality that is expected on
> one device
> (for some definition of "core" the WG will agree on later).
>
>
> Andy
>
> And in between there lies the case of needing to touch some, but not all
>> routers
>> because other routers in between are worked on by signaling/routing.
>>
>> So the answer is 1 <= #agents <= #routers
>>
>> And that answer is modulo there being only one agent on each router :-)
>>
>> For (crass) examples of the above I give you:
>>
>> Set up an LSP using RSVP-TE (just talk to head-end router)
>> Set up an end-to-end static route (talk to each router)
>> Set up a multi-segment pseudo-wire (talk to the T-PEs and the S-PEs)
>>
>> Adrian
>>
>>
>
> _______________________________________________
> i2rs mailing list
> [email protected]
> https://www.ietf.org/mailman/listinfo/i2rs
>
>
_______________________________________________
i2rs mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/i2rs

Reply via email to