[c-nsp] 7600 Owners, failure stats wanted

[Pete Templin]

On Fri, Jan 20, 2012 at 06:19, James Bensley<jwbensley at gmail.com>  wrote:

> I can have dual RSPs but how likely are they too fail? I want to know
> from 7600's owners/managers out there, how many SUPs or RSPs have you
> had fail on you (or not if non have failed on you), and how long were
> they in service before they failed (or how long were they in service
> without failing, if that is the case for you)?
>
> Are they more likely to fail the other line cards? It doesn't seem
> very common practice to have to of every card in the chassis and
> provide customers with a port on two switching modules for example, so
> why dual RSPs? Are they *that* much more likely to fail?

In a prior job in the SP world, our model was a hybrid:

"Router 1" got dual routing processors (RSP, SUP, GRP, whatever), 
"Router 2" got single RP.  Customers who cared would pay for two links 
and gain box-level redundancy, customers who didn't would be provisioned 
on router1 and at least have controller-level redundancy.  If router1 
would fill, the plan was to give router2 dual RP and deploy router3, 
though now I'm thinking it'd be smarter to deploy router3 with dual RP 
and leave router2 for just the customers with twinned links.

Our objective was to make a reasonable effort to keep the customers 
online, and if things went bump in the night that we'd wake up the 
fewest personnel possible.  It wasn't necessarily that an RP would die, 
but for some platforms the RP would crash and reboot.  We'd rather our 
customers go through an RP switchover than a hard-down router reboot. 
We didn't maintain two of everything - just a regional spare.

We gave up on dual-RSP 7507s for T1 aggregation and settled on the 
7206s; the price/volume point didn't make sense to choose a larger 
dual-RP platform for us.  We were headed towards GSR for higher-speed 
TDM access, and used the above model on 6509s for customer Ethernet access.

pt