Fabric design and deployment
Domain 1 checks whether you can turn requirements - sites, service counts, dual-homing, WAN reach - into a working Fabric Connect (SPB) design before a single CLI line is typed. Expect role assignment, B-VID and nickname planning, I-SID allocation, MTU budgeting, and decisions about where Fabric Extend replaces a native fabric link.
Roles: every node is a BCB, some are BEBs
- The fabric is IEEE 802.1aq Shortest Path Bridging run by a single IS-IS Level 1 instance. There is no spanning tree in the core, so every fabric link forwards.
- A node that only transits is a BCB; a node that also carries access ports or an L3 service is a BEB. Same software, different placement.
- Each node needs a unique 20-bit nickname. If you do not set one, the system derives it from the system-id - fine in a lab, risky in a planned fabric.
- Two B-VIDs back the fabric (4051/4052 is the common pair) so primary and secondary trees provide load sharing and repair.
Service plan
- The I-SID names a service end to end; VLANs are local. One I-SID per L2 VSN, one per L3 VSN (VRF), plus the management VSN at I-SID 16777215.
- Reserve blocks up front: management, VRFs, tenant L2 services. Document before deployment so ranges never collide.
- L3 VSN with distributed routing places the same gateway IP on multiple BEBs for first-hop optimisation.
MTU and WAN edges
- Fabric encapsulation adds overhead. Size the core for the largest frame you intend to carry and keep it uniform: an MTU mismatch appears as silent large-packet loss long after IS-IS is already up.
- Fabric Extend carries SPB over IP (UDP 8472), so a site behind an L3 WAN can still be a fabric node - at the MTU the WAN actually guarantees.
- SMLT with vIST dual-homes access switches and servers without a physical IST link.
Worked example
Two data centres, 60 access closets, one remote office reachable only over an MPLS L3 VPN. Design: two core BCBs per DC in a square mesh; each closet dual-homed to a BEB SMLT pair using vIST; the remote office is a single BEB with two Fabric Extend tunnels across the VPN. Nicknames 0.00.01.01–0.00.01.20 per site block; I-SIDs 10000–10999 for L2 VSNs, 20000+ for VRFs; management stays at 16777215. Because the MPLS path only guarantees 1500-byte frames, the FE tunnel MTU is set to 1400 and tenant MTU documented at 1400 - not the 9000 used inside the DC.
Traps that cost marks
Assuming the core still needs STP (it does not, though STP still matters on UNI access ports); reusing a nickname because "adjacency came up anyway" (it will flap and black-hole MACs); treating a VLAN ID as globally meaningful (only the I-SID is); forgetting 16777215 is the management VSN; designing a single B-VID, which removes the second tree and its redundancy.