Ultimately, many of the I2RS applications concern what you want to do with the network, and having a model that represents an abstraction of the network as a whole (as opposed to individual devices) seems to be core of what a controller needs to do. I think there is therefore no question that a topology model is needed. The question that is debatable concerns whether the interface that is exposed northbound of a controller (which will include a topology model) is within scope. Even if the interface is not in scope (there is certainly validity to the scope creep argument), retaining the model by itself in scope makes sense: as a "use case" and a source of requirements (as it needs populating), and as a way to capture topology information as it is understood by the devices themselves. --- Alex
-----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Eric Osborne (eosborne) Sent: Thursday, November 07, 2013 12:52 PM To: Joe Clarke (jclarke); Alia Atlas; [email protected] Subject: Re: [i2rs] topology info model - what makes it a "network" model vs. a "device" model If you want to get the link state for an entire network there are two ways to do it: 1) the i2rs client fetches each node's link state info from that node, and plugs them together to make a network, 2) the i2rs client fetches the entire LSDB from one node. IMO either one is OK, and #2 makes the most sense. In any case it's a read-only thing so it doesn't really matter how it gets done. Where it gets hairy is when you want to write to the network. The question that kicked off this thread was along those lines, if I recall. If you want to get a mutex on the network as a whole you *need* a way to lock all devices. Perhaps this is best done in the client - acquire a lock on the config for each node in the network, make your changes, then unlock. If there were a higher level way to do this with some sort of network API that's what it would have to do under the covers anyways. And if you have locking write access to the client then you don't need a network model, just a router model. eric > -----Original Message----- > From: Joe Clarke (jclarke) > Sent: Thursday, November 07, 2013 12:39 PM > To: Eric Osborne (eosborne); Alia Atlas; [email protected] > Subject: Re: [i2rs] topology info model - what makes it a "network" > model vs. a "device" model > > On 11/7/13, 12:11 AM, Eric Osborne (eosborne) wrote: > > If Not: what problems arise that would be solved by having a network > model, do we need one, and where would it come from if not i2rs? > > I think it is useful to have a network-wide topology model that is > accessible at a device level. Meaning, topology could be distributed > into the network and retrievable either as a whole network or in parts > based on filters. For example, given an I2RS Agent, give me the Agent > plus two routed hops. This would be very useful, for example, in the > support case where we need to understand what a network looks like in > order to help troubleshoot problems. > > But, I could also ask this same Agent to give me the whole routed > network from a given instance. This would give me the graph (edges, > nodes) of that topology. > > Joe > > -- > Joe Marcus Clarke, CCIE #5384, | | > SCJP, SCSA, SCNA, SCSECA, VCP ||||| ||||| > Distinguished Services Engineer ..:|||||||||::|||||||||:.. > Phone: +1 (919) 392-2867 c i s c o S y s t e m s > Email: [email protected] > > ---------------------------------------------------------------------- > -- > ---- _______________________________________________ i2rs mailing list [email protected] https://www.ietf.org/mailman/listinfo/i2rs _______________________________________________ i2rs mailing list [email protected] https://www.ietf.org/mailman/listinfo/i2rs
