[c-nsp] IOS XR

[Rob Shakir]

On 12 Sep 2011, at 11:45, Vitkovsky, Adam wrote:

> Well you can always pay some extra $$$ for the 1+1 APS to improve the convergence times on the long-haul systems
> But still if the first aplifier on the TX path fails you'd still have to wait some time till the blackout reaches the RX site at the oter end of Pacific
> Than you'd just need to relay on some increased error counts or decreased signal levels right before the aplifier fails -which you'd use to switch and start RX from the alternate working path assuming the 1+1 APS

This is one of the nice things about looking at convergence between the optical and IP/MPLS layers - as I think is outlined in [0], there's now some possibility for the IP/MPLS layer to get some insight into what's happening at the optical layer, and trigger protection on this basis. If one utilises the pre-FEC error count, then theoretically, if the protection mechanism is applied in the small window prior to the channel becoming completely unavailable (and rather whilst it's in some failure mode that the level of FEC is able to protect you from), then it's possible to avoid having a service interruption during such a failure.

To pull back to the original subject, I think Mikael's point here around what _exactly_ we define ISSU as is really important - whilst we've definitely seen some improvement in this over the last few years, a couple of seconds is still unacceptable in some of the networks that I'm working with. The problem is always going to be that engineering a box for <50msec ISSU performance is going to be expensive. In addition, it really increases the testing load that is required for QA of such a solution (as was pointed out in this thread - if it's unreliable, ISSU can leave your device in a lot more of a mess than a 'normal' upgrade, and hence result in a much longer outage than just reloading it). 

IMHO, there are definitely deployments that are willing to pay a premium for this behaviour, it's more a point of whether there are enough of these deployments to actually make the business case for delivering a <50msec ISSU solution viable.

Cheers,
r.


[0]: http://www.terena.org/activities/ngn-ws/ws2/pennell-ipodwdm.pdf