[c-nsp] Cisco BFD and NSF

[Saku Ytti]

On (2006-07-31 09:04 +1000), David J. Hughes wrote:
 
> Wow, that's a major disappointment.  So much for BFD being  
> implemented in the forwarding plane.

How would this work? If BFD was off-loaded to linecards? Linecards would
just send 'down event' to control-plane when BFD goes down in particular
linecard? And 'show bfd ...' would actually poll real status from each
line card every time executed?
 Doing this type of off-loading to me sounds bit risky, since the
BFD is running in different place than the routing-protocols. What
if linecard fail in manner which makes it not send the bfd 'down event'?
Routing protocols would keep on going until their build-in livenyless
detection detects problem, as they are not ran in the linecard?
 Obvious fix is to pass messages more or less as often between linecard
and control-plane as between linecard and network. But then, how is
this easier than passing directly from control-plane to network, without
offloading?

 To me doing this kind of guarantee in control-plane, given modern
software and non-ancient CPU sounds like extremely feasible, under
extreme loads. So perhaps this is just issue due to historical
reasons in IOS, perhaps this is already being fixed by ION stuff in
SRA?

-- 
  ++ytti