[c-nsp] svi autostate issue

Андрей Андреев a.andreev at teztour.com
Fri Dec 16 00:57:02 EST 2011


I mean indirect ring link failure.

Assume Switch A (not STP root) detect ring interface DOWN.
Switch A send TCN bpdu on path to root, when root receive TCN bpdu it 
send TCA bpdu to all ports.
Switch B (it has ospf and eBGP adjacencies) receive TCA bpdu and put all 
not edge stp port on BLK state.
And autostate feature put SVI in DOWN state.
Then it start proposal/agreement procces to select new root port.
When the process finish Switch B unblock ports.





15.12.2011 21:45, Peter Rathlev пишет:
> On Thu, 2011-12-15 at 16:41 +0400, Андрей Андреев wrote:
>> Some SVI has OSPF adjacencies and some eBGP to clients.
>> In period of STP convergence SVI flaps and OSPF/BGP flaps too.
> And your STP convergence means that the line-protocol flaps, right? So
> it's not the convergence time as such, but simply the fact that the
> line-protocol flaps and that this is signalled to the routing processes.
>
> I don't think there's a way to tell OSPF to wait for the usual timeouts
> instead of reacting to link down, it's a link-state protocol after all.
> But OSPF convergence can be fast too, so my main concern would by CPU
> load from the churn.
>
> I don't understand why eBGP would flap though. AFAIK, unless you enable
> "neighbor n.n.n.n fall-over", it shouldn't tear down an eBGP session
> just becase the line-protocol drops.
>
> A dirty hack would be to use two physical interfaces and a looping cable
> to always carry the VLAN(s).
>
> If per-VLAN STP was a possibility, topology changes would at least only
> affect the VLANs where it was relevant.
>
> Or ask Cisco to implement "no autostate" on the 6500. :-)
>



More information about the cisco-nsp mailing list