cancel
Showing results for 
Search instead for 
Did you mean: 

NNI as a DCI - Best practice

NNI as a DCI - Best practice

AlexT0
New Contributor
Does Extreme have any best practices for DCI links configured as NNI's for a pair of 8404's?

These NNI's will be be connecting BCB's, So at this point of time they are strictly just transit nodes and are not doing any L3 features for the network, VRRP, RSMLT, etc.

My concerns were more along the lines of IS-IS path computation / re-convergence, If we had a carrier link flap would we want to dampen this with something like VLACP?

Additionally would we want to implement VLACP for a quicker link detection failure if the link was to remain up, but the carrier was not allowing traffic through their network, or would IS-IS hellos detect this is a quick enough manner?
1 ACCEPTED SOLUTION

Roger_Lapuh
Contributor II
AlexT0,

Recalculation of paths is typically done in a few 10s-100s of milliseconds in an SPB network, depending on number of nodes and paths in your area. Typically, the biggest impact to convergence is how long it takes to detect a link down.

Everything depends on your DCI connection type:

If you have just dark fiber or xWDM, then you can rely on Ethernet to detect link failures. Failures are typically detected in10-20 milliseconds in that case by the Ethernet interfaces.

If the underlying DCI connection is an L2 pseudo-wire service from a provider with active components, then yes, it does make sense to use VLACP on those links. (there is also a flap dampening feature available)

If the underlying DCI connection is an L3 routed service (IP VPN/MPLS) and you are using Fabric Extend IP, then you want to detect a failure using BFD between the FE tunnel end-points.

If you are concerned about flapping of links, we recently introduced a hold-up timer for Ethernet ports, with this you can keep the interface up even though there might have been a blip on the carrier side, this can be beneficial in some deployments.

IS-IS by default takes 27 seconds to detect a link down, so it is not fast and should not be used to tweak convergence times. Use above methods instead.

Roger

View solution in original post

2 REPLIES 2

Roger_Lapuh
Contributor II
AlexT0,

Recalculation of paths is typically done in a few 10s-100s of milliseconds in an SPB network, depending on number of nodes and paths in your area. Typically, the biggest impact to convergence is how long it takes to detect a link down.

Everything depends on your DCI connection type:

If you have just dark fiber or xWDM, then you can rely on Ethernet to detect link failures. Failures are typically detected in10-20 milliseconds in that case by the Ethernet interfaces.

If the underlying DCI connection is an L2 pseudo-wire service from a provider with active components, then yes, it does make sense to use VLACP on those links. (there is also a flap dampening feature available)

If the underlying DCI connection is an L3 routed service (IP VPN/MPLS) and you are using Fabric Extend IP, then you want to detect a failure using BFD between the FE tunnel end-points.

If you are concerned about flapping of links, we recently introduced a hold-up timer for Ethernet ports, with this you can keep the interface up even though there might have been a blip on the carrier side, this can be beneficial in some deployments.

IS-IS by default takes 27 seconds to detect a link down, so it is not fast and should not be used to tweak convergence times. Use above methods instead.

Roger

Understood, looks like my initial assumption was correct and VLACP would be a good idea here.

Thanks once again sir,
GTM-P2G8KFN