[j-nsp] eBGP neighbor link failure detection

[Vitkovský Adam]

You can verify whether the fast external failover is enabled for the peer/globally (should be on by default). 
-in that case the 3min hold down timer does not need to expire when the link to neighbor fails and the adj-rib-in is flushed immediately. 

Also you can verify whether the BGP Next-Hop Address Tracing is enabled (should be on by default). 
When enabled BGP does not have to rely on the Scanner process running every 60sec to detect that the NH for the prefixes is unreachable. 
With Next-Hop Address Tracing BGP registers NHs with RIB process which notifies BGP (delay can be tuned) each time status/metric of the prefix corresponding to NH changes. 

The above are techniques local to a BGP speaker. 
The other aspect is the propagation of BGP updates throughout the AS. 
To eliminate the propagation delay there are features like local-repair or PIC (also local a BGP speaker with a failed link). 

adam