[c-nsp] BGP Next-hope convergance

[Mikael Abrahamsson]

On Mon, 8 Sep 2008, Chintan Shah wrote:

>  Hi Guys,
>
>   I have 6509 (PE3) switch learning VPNV4 routes from two PE1 & PE2 and
> select
>   best path PE1 ( based on IGP metric).
>   Now, PE-1 uplink to core failes and CR(P) find that PE1 not reachable
> based
>   on IGP dead time = 9 sec and PEe learn that PE1 no more reachable now (
>   another 6 sec) so totatl IGP convegance =15 sec. howerver PE3 take
> another
>   5-6 sec to select best path is PE2 now for remote VPNV4 routes and total
> i
>   get 21 sec....Why PE3 doesn't select PE2 afer 15 sec (IGP convergance)
> and
>   wait another 5-6 ...what is that 5-6 can cause overvall convergance 21
> sec
>   ??
>   Can some one help me to understand ?

First of all, what are your RD values? You need to make sure RD values are 
unique for each PE (use the IP:value format, instead of the AS:value 
format commonly used).

Also, do the VPNv4 BGP sessions go down promptly when the next-hop becomes 
unreachable? If the platform supports BFD, you should use that to make the 
IGP session go down quicker.

An easy way around this is also to make sure PE1-PE2 are connected 
somehow, so a PE never becomes unreachable due to a single link failing.

With proper design you will be able to reach 0-2 second convergence 
time...

-- 
Mikael Abrahamsson    email: swmike at swm.pp.se