MLAG Peer-Link Saturation During VSU Uplink Failure
A two-switch campus core runs MLAG between Core-A and Core-B over a 2x40G peer link (LACP bond, two members). Access closets dual-home via MLAG to both cores. The peer link carries both MLAG control traffic and the dual-homed access traffic that must traverse it when a downstream uplink fails. On Tuesday, Access-SW1's uplink to Core-A fails. Within 90 seconds, users behind Access-SW1 report severe latency and some VoIP calls drop. Monitoring shows the peer link at 97% utilization and the MLAG control channel logging heartbeat delays. Core-A and Core-B remain MLAG-primary/secondary and the VLANs stay up. No spanning-tree topology change was recorded on the access switch.
1. Diagnose why the MLAG peer link saturated and explain why the failure did not trigger a spanning-tree reconvergence that would have shed load.
Read the worked answer
When Access-SW1's link to Core-A fails, all of its dual-homed traffic that was hashed across both cores must now reach Core-B via Core-A across the MLAG peer link. MLAG is a control-plane abstraction: both cores present as one logical LAG endpoint, and the peer link is the data path for any traffic that ingresses one peer and must egress the other peer's downlinks. Because MLAG keeps both peers forwarding (no STP blocking on the dual-homed access links), there is no STP topology change to reconverge and relieve the peer link. The peer link therefore becomes a transit path and saturates. The heartbeat delays are a symptom of control traffic sharing the same oversubscribed link, not the cause.
Key judgement: the peer link is a transit path in failure scenarios, so it must be sized for worst-case single-uplink-failure traffic, not just steady-state control traffic.
2. Recommend a design change and an operational safeguard. Justify each against the observed numbers.
Read the worked answer
Design: increase the peer-link capacity so worst-case transit fits with headroom. If steady state was ~50% of 80G and a single uplink failure pushed it to 97%, the transit load is roughly the failed uplink's share. Adding peer-link members (e.g., to 4x40G) or moving to higher-speed members restores headroom. Alternatively, reduce transit exposure by ensuring access dual-homing is symmetric so a single uplink failure shifts a bounded amount of traffic.
Operational safeguard: separate MLAG control traffic from the saturated data path where the platform supports a dedicated control path or QoS prioritization, so heartbeats are not starved. Also monitor peer-link utilization with an alert threshold well below saturation (e.g., 70%) and alarm on MLAG heartbeat delays. Validate the change with a controlled uplink-failure test measuring peer-link utilization and call quality.