Light ENNI – aka Split LAG … Region C 3 b b Region A c Link Aggregation Basic Forwarding Rules : a frame from the network is loaded into a locally-attached LAG link : if the whole LAG on that node fails, it blocks its “Region-facing” ports on egress from a LAG, a frame may have to pass over an IST (==) to ensure that it enters the destination region via the correct port : only one port can be open to any destination, otherwise frame looping and duplication start, and MAC learning breaks; MAC learning is also shared and synchronised over the IST ISTs must be proof against single point of failure :-“split brain” issue : modulo Norm’s “suicide pact” suggestion; they can also be built as LAGs in their own right if desired

Light ENNI – aka Split LAG … Region C 3 b b LAG Region A c Link Aggregation The easiest way to access this structure is “triangular” Split LAG : exactly one frame is sent in, and exactly one frame comes out But this looks very “heavy” for a light ENNI  The alternative is to make {a, b, c …} full participants in their region’s control regimes : which if this is xSTP looks “interesting”; next slide

Light ENNI – aka Split LAG … Region C 3 b b Region A c Link Aggregation The alternative is to make {a, b, c …} full participants in their region’s CP : xSTP is the challenging case There can only be a single unblocked point of attachment of the virtual bridge (3a, 3b, 3c ) to the region’s Spanning Tree on any VLAN : the structure above duplicates frames on flooding , and “thrashes” source learning for frames travelling  The Spanning Tree root cannot be placed “on” a virtual bridge element : the Region may have > 1 E-NNI

Light ENNI – aka Split LAG … PN Region C 3 b b Region A c Link Aggregation In a Link State environment, the solution is straightforward : all bridges in Region C can obtain a common view of which of {a, b, c} is on the shortest path to a “PN”, and act accordingly When running xSTP, the ENNI probably has to do this “by the back door” : the xSTP on {a, b, c} has to be extended to allow “private” sharing of Root Path Priority Vector between a  b, a  c, and b  c : with the winner being the owner of the unblocked port; unless there is a better idea … ?

“One we made earlier”

SPBM for resilient PB E-NNI NW 1 PB NW 2 S-Tag = x S-Tag = y I-SID = z Operators 1 & 2 bind each PB S-tag to a common SPBM I-SID Each operator is responsible for advertising that I-SID once only into the SPBM domain : but each operator has complete authority as to where that is So long as a viable path remains available in the SPBM domain, connectivity will automatically be maintained by SPBM. I-SiDs are provisioned in SPBM : a signalled interface might be desirable MVRP ?