the spines have BGP peering, yes. Spine1 is getting the default route from the up-link, so that if the link from Spine2 to the VC fails, it still has connection. I understand the concept of load-balancing on layer3 with multipath, but is it somehow possible to do it on layer2 like with a classic virtual-chassis too?
------------------------------
Klaus Peter Schnarch
------------------------------
Original Message:
Sent: 03-03-2025 03:39
From: One
Subject: EVPN/VXLAN Layer2 load-balancing across two spines
I'm not sure I understand your topology from the description so I'm guessing the answer should be that this is possible, but preferably not with LACP.
The spines should be running BGP and have 1 IP-link each toward the VC, if the VC is intended to be leafs in this fabric. You peer both spines with the VC and do per packet load balancing as a load-balancing policy.
set policy-options policy-statement LOAD_BALANCE term LOADBAL then load-balance per-packet
set policy-options policy-statement LOAD_BALANCE term LOADBAL then accept
set routing-options forwarding-table export LOAD_BALANCE
- If the VC is not running BGP/EVPN/VXLAN themselves, they should probably be connected to a pair of leafs instead of the Spines. In this case you deploy LACP with ESI. You can probably do this on the Spines as well, but not recommended in traditional CLOS fabric design.
- It is not recommended to have a direct interconnect between spines in an EVPN/VXLAN fabric. Traffic between them should traverse the leafs.
EDIT:
I now see the diagram, was not included in mail summary I read initially.
In this case - I still don't understand the Spine layer, what protocols are they running? Are they BGP-peering directly and is Spine1 getting a route-advertisement about the networks reachable on the up-link (default route)?
Original Message:
Sent: 03-01-2025 15:22
From: Klaus Peter Schnarch
Subject: EVPN/VXLAN Layer2 load-balancing across two spines
Hello,
we have two spines, which are interconnected with 100G link and a downlink on each spine (40G) towards a virtual-chassis. The VC is configured as LACP trunk (80G in total) and is in the same ESI on both spines.
Now there is one single 100G Uplink on one of the spines. Is it possible to load-balance the layer2 packets across both spines to the VC? I mean just like with a virtual-chassis, where both devices bandwidth can be used together with LACP. So that in the end 80G can be forwarded towards the VC
------------------------------
Klaus Peter Schnarch
------------------------------