On Tue, Nov 05, 2013 at 02:39:09PM +0000, Acee Lindem wrote: > Hi Mark, > I attended and the majority of the discussion centered on whether the > problem could be solved with a simpler model such as a FIB per provider. > Fred pointed out that this would not handle overlapping source subnets and > the fact that it was possible to solve the general problem with a Patricia > tree and the installation disambiguating routes. Alvaro also commented > that there are use cases beyond homenet. As you'd expect, the next step > was that there needs to be more discussion on the RTG WG list (copied).
I disagree that it needs more discussion, at least on this particular item. It's an implementation detail. The way it's currently described does somewhat imply single-FIB, but it doesn't prescribe it, and in the end it *is* the match behaviour we want - so, I'd just interpret it as requirement on the _behaviour_ and if anyone wants to go with multiple FIBs in their implementation, feel free... (Yes, trying to put the "implementation detail" lid on the discussion here.) If anyone thinks we should have different match behaviour, that's a different discussion (but I haven't heard anything in that direction.) Describing the same match behaviour with multiple FIBs per source doesn't sound easier to me. I believe the current description is as simple and down to it as it gets. -David P.S.: I'm not even sure if a "FIB per provider" model is actually simpler, and I'm pretty certain the implementation is more complex. You'd have to copy most of the FIB content to each of the FIBs, dynamically create FIBs when upstreams appear/disappear (on remote routers even), and then there's also the case of an upstream going down => its addresses being deprecated => host still using them for old local connections => router should retain local connectivity for deprecated prefix => router needs to retain FIB with timer. Also needs to be done with routes, but I'd prefer doing it on routes only, and not on routes and FIBs. And I almost deleted that PS - really, implementation detail! > Thanks, > Acee > > On 11/4/13 11:05 PM, "Mark Townsley" <[email protected]> wrote: > > >Was anyone on this list able to attend the rtgwg meeting today where Fred > >presented draft-baker-rtgwg-src-dst-routing-use-cases-00? I missed it, > >and would be interested in the reaction, feedback, or next steps (if any). _______________________________________________ rtgwg mailing list [email protected] https://www.ietf.org/mailman/listinfo/rtgwg
