Hi Peter, > The version -04 does not contain normative MUST that UPA shall only be > > used to trigger invalidation when end to end encapsulation is used for > > subject application(s). So as written is in fact quite undeployable in a > > mixed vendor and legacy node(s) environment doing hop by hop routing. We > > can't just hope that this is all about configuring the network in a > > proper way. > > above looks more like a comment that can easily be addressed by > additional text, which I'm willing to add, rather than an objection to > draft becoming a WG document. Please let me know if you think otherwise. >
Ok if you are willing to add such restriction to the spec then I can remove my concern #1. > The solution is too pragmatic ... > > Isn't that a good thing, isn't it? :) > :) Yes and it does assure that we can have some mechanism like this before we all retire ! But you know once you have an interim solution it is much harder to convince folks to work on a more optimal one. > It does not look at the problem > > holistically. Yes I still think the problem is worth solving but outside > > of link state IGP doing UPA blast flooding everywhere domain wide even > > if no nodes need that info. As discussed at length it could be done via > > either BGP indirection or via PUB-SUB model (as proposed). > > I don't object to solve this problem in BGP, or elsewhere. But given > that the problem we are trying to solve is created by the IGP > summarization and the ultimate source of the prefix reachability or > unreachibility comes from IGPs, having a solution in IGPs seems like a > natural choice. That is all correct. But I really do not like to use domain wide flooding for it. For example if you would contain the solution within IGP but from local or remote ABRs unicast by UDP to subscribers that given PE or POP is down that would also contain the solution to IGP. Yes it is a bit more difficult as you would need to solve subscription problem first as opposed to the blast or spray approach, but the current draft does not even leave any room for such apparatus. So my concern here is not that the proposal is broken. It is just suboptimal IMHO. And as we spoke a few weeks back in SF I think it is harmless, but is this enough ? Lastly, I think PEs do not accidentally go down that often, and for any planned maintenance there are other ways to drain the box. With that said, consider my objection to be a soft one based on the above. Cheers, R.
_______________________________________________ Lsr mailing list [email protected] https://www.ietf.org/mailman/listinfo/lsr
