Re: [Idr] draft-uttaro-idr-bgp-persistence-00

[<[e-mail]>]

Robert,

>Bruno,
>
>> The low chance / probability of the failure of both RR planes is
indeed
>> a good point. I can also agree that this is expected to be low. I
>> disagree that it never happens. E.g. with current BGP error handling,
>> the same BGP attribute in "error" would cause both iBGP plane to go
>> down. Hopefully work is being done to improve this. But I would not
bet
>> that this will cover all future cases.
>
>If trigger is due to malformed update why are we so concerned about
RRs?

We are concerned about RR-PE sessions because if the sessions are down,
all routes are down. PEs become essentially useless.

>Wouldn't equally susceptible be also PEs/ASBRs ?

Indeed, inter AS VPN eBGP session are also important. (although you
would one use the inter AS routes and hence PE/VPN are still useable.

>And the cure that those is GR, not persistence ;).

Cf Enke's thread.

>> In such catastrophic cases, some routing is better than no routing.
>
>I don't know about that.
>
>> On a side note, considering that the RR and hence their failure are
>> independent and therefore will quite never fail at the same time, is
>> questionable. To start with, they process they mostly process the
same
>> input (BGP updates) and are part of the same network.
>
>I don't think I used the term "never". 

Ok.

>But with dual or triple vendor
>control plane the probability of RR network wide outage drops
>significantly lower.

Sure. But given the small number of RR boxes, and the high requirements
on these boxes, it's harder to justify a two or three vendor sourcing.

Cheers,
Bruno

>Cheers,
>R.