Deployment Guide

14 Leaf-Spine Deployment and Best Practices Guide | Version 1.0
Note: Downstream connections from leaf switches configured for VLT do not necessarily have to be
configured as LAGs if other fault tolerant methods are preferred (e.g. multipath IO). In this guide, examples 1
and 2 use LAGs to downstream servers while examples 3 and 4 do not.
5.2 UFD
If a leaf switch loses all connectivity to the spine layer, by default the attached hosts continue to send traffic to
that leaf without a direct path to the destination. The VLTi link to the peer leaf switch handles traffic during
such a network outage, but this is not considered a best practice.
Dell EMC recommends enabling UFD, which detects the loss of upstream connectivity. An uplink-state group
is configured on each leaf switch, which creates an association between the uplinks to the spines and the
downlink interfaces.
In the event all uplinks fail on a switch, UFD automatically shuts down the downstream interfaces. This
propagates to the hosts attached to the leaf switch. The host then uses its link to the remaining switch to
continue sending traffic across the leaf-spine network.
5.3 RSTP
As a precautionary measure, Dell EMC recommends enabling RSTP on all switches that have layer 2
interfaces. Because VLT environments are loop-free, simultaneously running spanning tree is optional though
considered a best practice in case of switch misconfiguration or improperly connected cables. In properly
configured and connected leaf-spine networks, there are no ports blocked by spanning tree.
5.4 Routing protocols
Any of the following routing protocols may be used on layer 3 connections when designing a leaf-spine
network:
• BGP
• OSPF
• Intermediate System to Intermediate System (IS-IS)
5.4.1 BGP
BGP may be selected for scalability and is well suited for very large networks. BGP can be configured as
External BGP (EBGP) to route between autonomous systems or Internal BGP (IBGP) to route within a single
autonomous system.
Layer 3 leaf-spine networks use ECMP routing. EBGP and IBGP handle ECMP differently. By default, EBGP
supports ECMP without any adjustments. IBGP requires a BGP route reflector and the use of the AddPath
feature to fully support ECMP. To keep configuration complexity to a minimum, Dell EMC recommends EBGP
in leaf-spine fabric deployments.
BGP tracks IP reachability to the peer remote address and the peer local address. Whenever either address
becomes unreachable, BGP brings down the session with the peer. To ensure fast convergence with BGP,