Full mesh or H-VPLS for a 24-site retail VPLS
A regional retailer is replacing a hub-and-spoke Epipe design that forces all store-to-store traffic through two data-center PEs; measured round-trip latency between two stores in the same city is 38 ms. The MPLS core has six Nokia SR OS PEs and 10 Gb/s links; stores connect at 1 Gb/s. The retailer wants one VPLS broadcast domain for POS and back-office, with about 1,500 MAC addresses today and a forecast of 4,000 in three years. The operations team is uneasy about a full mesh: they have heard that VPLS uses split horizon and that broadcast traffic will multiply. They also need to choose between LDP and BGP-AD for pseudowire signaling. One option is a full mesh of six PEs; another is H-VPLS with two hub PEs in the data centers and four spoke PEs. No spanning tree is currently run at stores, and the customer switches are unmanaged.
1. Recommend a topology for the six-PE VPLS. Justify it using pseudowire count, forwarding path, and the three-year growth forecast.
Read the worked answer
At six PEs, a full mesh is 6×5/2 = 15 mesh SDPs. H-VPLS with two hubs is 4×2 = 8 spoke SDPs plus one hub-to-hub mesh = 9 pseudowires, but store-to-store traffic between two spoke PEs must go spoke→hub→spoke, adding latency and loading the hubs. With only six PEs and a 10 Gb/s core, 15 PWs is modest; 4,000 MACs is also modest. I would recommend full mesh now for direct any-to-any forwarding, and set a trigger, for example more than 10 PEs or hub utilization above 70%, to move to H-VPLS. Use BGP-AD for auto-discovery and route-target-based membership, or LDP if the provider wants simpler operations and manual SDP provisioning. The decision should be based on PW count, forwarding path, and operational model, not MAC scale alone.
2. Explain how VPLS split horizon affects a full-mesh design and whether spanning tree is still needed at the stores.
Read the worked answer
A frame received on a mesh SDP is not forwarded to another mesh SDP. That prevents a loop across the full mesh, but it also means a PE cannot use another PE as a transit point between mesh SDPs; every PE must have a direct mesh SDP to every other PE in the VPLS. Split horizon does not protect against loops in the customer access. With unmanaged customer switches and no STP, a physical loop at a store can still flood the VPLS. Recommend enabling STP or MSTP on the PE SAPs toward customer switches, or requiring the customer to run a loop-free topology. Add discard-unknown and per-SAP MAC limits as guardrails, and monitor FDB usage and broadcast rates.
3. Compare LDP and BGP-AD for this deployment. What would you verify on each PE?
Read the worked answer
LDP-signaled VPLS is simpler: the provider configures SDPs and mesh bindings manually. BGP-AD automates discovery using the VPLS-ID, RD, and route targets; it scales better but adds BGP policy and RT coordination. For six PEs, either works, but BGP-AD reduces provisioning errors as sites are added. Verify that the VPLS-ID and route-target import/export match on all PEs, that the RD is unique per PE, and that BGP VPLS routes are present. Check show service id <id> bgp-ad for remote peers and show service id <id> fdb for MAC learning. If LDP is used, verify the SDP far-end and VC labels instead.