Apologies ... want to correct myself for clarity:

was: "active and backup paths all going to the next hop X"

should be: "active paths all going to the next hop X and backup paths going
to different next hops"



On Thu, Jan 6, 2022 at 12:31 PM Robert Raszuk <[email protected]> wrote:

> Peter,
>
> So far you and others have been saying all along that one of the
> applications which uses PULSE can be BGP.
>
> If so I am afraid this is not PIC (== Prefix *Independent *Convergence)
> anymore. And I think this was Bruno's valid point.
>
> Today BGP registers with RIB next hops for tracking, When next hop goes
> down or metric changes BGP get's notification. So now there needs to be a
> new construct stating that registered next hop went away based on PULSE
> signaling and *all* best paths need to be recomputed (or in case of
> multipath enabled - removed). So yes this is a trigger, but it is *PER
> PREFIX*.
>
> That is my current understanding of how PULSE will be used. All of this is
> a guess as you keep this part as an implementation detail/secret :). That
> is also compatible with the notion of recomputation of the removed best
> path when PULSE expires.
>
> Now one could potentially do it in a different way - PIC style - that is
> first to preinstall in FIB active and backup paths all going to the next
> hop X and simply invalidate this next hop X when PULSE arrives with single
> adj rewrite - directly in the FIB itself.
>
> But in this case PULSE will never be heard by BGP or for that matter by
> any other app - not giving them a chance to select another best path in
> case of no iBGP multipath to be enabled. You could also propagate this to
> both BGP and FIB.
>
> What seems big unknown in the latter case is the operation aspect of those
> events in the light of the ephemeral nature of PULSE. I am sure you are
> going in either case to generate a syslog message upon PULSE reception.
> Extending current RIB, building shadow ephemeral RIB or bypassing it to
> talk directly between IGP and FIB is another operational concern.
>
> Thx,
> Robert.
>
> PS. Also keep in mind that in SRv6 used say for VPN applications you do
> not want to signal /128 next hops but SID locator prefix of the remote
> egress PE :) This is going to be fun !
>
>
>
> On Thu, Jan 6, 2022 at 10:25 AM Peter Psenak <[email protected]> wrote:
>
>> Bruno,
>>
>> the PIC is used unchanged with PULSE.
>>
>> The only difference is that the PIC is triggered by the pulse arrival,
>> instead of the IGP route removal. We have made a prototype of it and it
>> works fine.
>>
>> thanks,
>> Peter
>>
>
>>
_______________________________________________
Lsr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/lsr

Reply via email to