SMLT and vIST: Dual-Homing a Fabric Edge

How Split MultiLink Trunking dual-homes an edge device to a pair of switches with active-active links and no spanning tree, how virtual IST runs the inter-switch trunk through the SPBM fabric itself, and the smlt-peer-system-id and smlt-virtual-bmac that make the cluster one logical node.

Dual-homing without spanning tree

An access switch that uplinks to two upstream switches for redundancy traditionally depends on spanning tree, which blocks one of the two paths - you lose half the bandwidth and failover is slow. SMLT (Split MultiLink Trunking) takes a different route: the access device dual-homes to a pair of switches through a single logical link aggregation group, so both uplinks forward at once and failover is sub-second. When SMLT is enabled on a link aggregation group (an MLT), spanning tree is automatically disabled on it.

The switch cluster and the IST

The two upstream switches form a cluster and must stay in sync - MAC tables, ARP, and so on - over an Inter-Switch Trunk (IST). You configure an MLT of type SMLT with the same MLT ID on both peers, and the access device below sees one logical aggregation group rather than two separate uplinks.

Virtual IST

Traditionally the IST is a dedicated physical link between the two cluster switches. Virtual IST (vIST) runs that trunk as a channel through the SPBM fabric instead, so no direct cable between the peers is required. You create a vIST VLAN with an I-SID and a Layer 3 interface, then point the peers at each other:

vlan create 4053 type port-mstprstp 1
vlan i-sid 4053 40534053
interface vlan 4053
  ip address 10.0.53.1 255.255.255.252
exit
virtual-ist peer-ip 10.0.53.2 vlan 4053

Telling the fabric about the pair

For SPBM to interoperate with SMLT, each peer needs two settings under router isis:

spbm 1 smlt-peer-system-id 020b.0002.0000
spbm 1 smlt-virtual-bmac 02:0b:00:01:00:01

The smlt-peer-system-id is the system-id of the other peer, so that if it fails the local switch can forward on its behalf. The smlt-virtual-bmac is a backbone MAC that both peers share and advertise, so the rest of the fabric sees the cluster as one logical source for dual-homed traffic. The virtual B-MAC is only required when you use custom system-ids; with the default base-MAC system-ids the cluster simply uses the lower of the two. It must be unique in the fabric, and the common convention is to reuse one node's B-MAC and change the last byte. Roles are automatic: between the two peers, the lower system-id is primary.

RSMLT, SLPP, and where it fits

RSMLT (Routed SMLT) extends the same idea to Layer 3, adding routing redundancy across the pair; it requires a vIST. To guard against loops, VOSS runs SLPP (Simple Loop Prevention Protocol) on the cluster, and on an EXOS Fabric Attach edge you enable SLPP-Guard on the access ports - never on the uplinks. Put together, a vIST core pair is the standard way to build a redundant edge for a Fabric Connect campus: dual-home your access (including an EXOS edge over Fabric Attach) to the cluster, run the inter-switch trunk over the fabric itself, and let SMLT give you active-active links with sub-second failover. The fabric-identifier tool will decode the system-ids and virtual B-MACs you configure here.