RFC 7432 introduced the modulo-based Designated Forwarder (DF) election algorithm, and RFC 8584 improved it with the Highest Random Weight (HRW) algorithm. Together, these methods offer an efficient, automated solution for DF selection. Despite these advancements, there are operational circumstances where automation alone is insufficient. For example, during planned maintenance, network upgrades, or troubleshooting, operators may need to override automated elections to direct traffic in a controlled manner. Recognizing this need, RFC 9785 introduces the concept of preference-based DF selection. This method empowers administrators to assign explicit preferences to devices, ensuring that DF roles align with operational requirements. This solution reuses and extends the DF Election Extended Community defined in RFC 8584, which is advertised along with the EVPN RT-4.
RFC 7432 specifies six approaches for generating an Ethernet Segment Identifier (ESI). In our test, we focused on type 1, which leverages the LACP system MAC address and Port Key to derive the ESI. This automated process ensures each ESI is unique, eliminates the need for manual configuration, and reduces the risk of errors.

In the SR-MPLS testbed, we tested these two features separately. Firstly, we tested preference-based DF. We sent bi-directional broadcast traffic to verify that the DF is the only transmission point in the multi-homed group. Then we performed a preference change to trigger the switch over in the multi-homed group. During the test, we observed no packet loss or duplication. And the DF election followed the highest-preference algorithm.

Next, we conducted an EVPN auto-ESI test in the SR-MPLS test bed. We confirmed that the ESI was automatically generated using the LACP system MAC address and port key, as specified in RFC 7432. By examining the BGP EVPN Route Type 4 (RT-4) advertisements, we observed that both PEs had the same ESI, confirming their membership in the same multi-homed group. Due to time constraints, we were unable to send test traffic.

We have also performed the same test case under the EVPN-VXLAN testbed.

In the first test run, we configured the preference-based DF election together with auto-ESI derivation. We sent end-to-end unicast traffic using Keysight IxNetwork and observed zero packet loss. We then modified the preference values to trigger a DF switchover and confirmed that the DF role successfully transitioned to the highest-preference PE.

In the second run, we reconfigured the same Ethernet Segment to use the traditional modulo-based DF election algorithm as a baseline for comparison. Unicast traffic sourced from the ES was correctly forwarded exclusively through the elected DF PE, and the DF election functioned as expected under the modulo method, providing automatic but non-deterministic DF election across the participating PEs.