[j-nsp] MX960 Redundant RE problem

[Joel jaeggli]

On 2/15/12 10:56 , Daniel Roesen wrote:
> On Wed, Feb 15, 2012 at 12:24:50PM -0500, Stefan Fouant wrote:
>> The cool thing is the Backup RE is actually listening to all the
>> control plane messages coming on fxp1 destined for the Master RE
>> and formulating it's own decisions, running its own Dijkstra,
>> BGP Path Selection, etc. This is a preferred approach as opposed
>> to simply mirroring routing state from the Primary to the Backup
>> is because it eliminates fate sharing where there may be a bug
>> on the Primary RE, we don't want to create a carbon copy of that
>> on the Backup.
> 
> I don't really buy that argument. Running the same code with the same
> algorithm against the same data usually leads to the same results.
> You'll get full bug redundancy - I'd expect RE crashing simultaneously.
> Did NSR protect from any of the recent BGP bugs?
> 
> The advantage I see are less impacting failovers in case of a) hardware
> failures of active RE, or b) data structure corruption happening on both
> REs [same code => same bugs], but eventually leading to a crash of the
> active RE sooner than on the backup RE, or c) race conditions being
> triggered sufficiently differently timing-wise so only active RE
> crashes.

when ISSU actually works it's a godsend.

> Am I missing something?
> 
> Best regards,
> Daniel
>