Re: [OSPF] Poll for WG adoption of draft-hegde-ospf-node-admin-tag

[Rob Shakir <[e-mail]>]

Hi Peter.

On 26 Aug 2014, at 16:43, Peter Psenak <[e-mail]> wrote:

> 
> On 8/26/14 17:32 , Hannes Gredler wrote:
>> 
>> operators want to assign node-tags as per router function (ABR, PE, core) and then
>> the LFA-selection becomes much easier to specify. - e.g.
>> - only pick a LFA that does not cross another PE router.
>> 
>> similarily it is desirable for "LFA tunnel termination"
>> to put out a constraint which says
>> - only pick a PQ neighbor which has node tag 'X'
> 
> my point is that with the above approach you have to:
> 1. On candidate PQ nodes configure the tag X
> 2. on all other nodes configure "only pick a PQ neighbor which has node tag 'X'"
> 
> It's (2) which makes me feel uncomfortable, as it's a config to be applied to many nodes.

I’m unclear on how one would solve this — the key thing is that there are number of scenarios where it is *operator* preference rather than node capabilities that mean that we want to select a particular node for some certain application. This preference may be on a per-calculating node basis. If this is the case, then a single capability that says that a particular target node is capable of acting in a particular role is not sufficient.

(Consider this scenario:
	- rtr-A is in country 1.
	- rtr-B is in country 2.
	- Both rtr-A and rtf-B are capable of acting as PQ nodes,and need to act as such for ‘local’ nodes (i.e., those in the same country as them).
	- rtr-A should never select rtr-B as a PQ.
In this case, we need some tag that specifies country, as well as some tag that specifies that it is a valid PQ node. We then need specific policy on rtr-A and rtr-B to implement this policy.)

It is very typical that where we have such policy implementations, then we need to configure the behaviour on a per-node basis. This is especially true where policies must consider characteristics of the topology.


>> 
>> i found it always strange that we for TE (as an example for
>> constraining paths) we have got ways to tag links, but
>> not way to tag nodes - that draft aims to fix that.
> 
> I'm not against tagging nodes as such. What worries me if we end up using node tags for signalling capabilities of node.

As per the above, I do not think that this mechanism replaces any capability, it just gives an operator a means to be more granular than the binary “supported”/“not supported” view that a flag indicating capabilities does.

I, of course, support the adoption of this draft as a co-author.

Cheers,
r.