Dual-Homed Server LACP Skew and Timer Change
Two VSP 7400 switches, A and B, form an MLAG pair. The peer link is a 2x40G LAG. A database server is dual-homed with a 2x25G LACP bond to A:1:25 and B:2:25. The server admin reports that one link is at 90% utilization and the other at 20%. The bond uses LACP active with long (90 s) timers. Last week one member flapped for 45 seconds during a switch reboot. 'show mlag' shows peer state FULL; 'show lacp' shows both links 'Selected'; 'show lag' shows the LAG up. The server has eight long-lived TCP flows. The team wants to 'fix LACP' by changing the timer to fast and adding a third 25G link. Hashes on L2 and L3. No configuration changes have been made yet.
1. Is the uneven utilization evidence of an MLAG or LACP fault? Explain.
Read the worked answer
No. MLAG peer is FULL and LACP shows both members Selected, so the control plane is doing its job. LACP does not stripe packets per link; the switch hashes each flow over the members. Eight long-lived TCP flows can easily hash to a small subset of links, producing 90/20. That is normal statistical skew, not a failure. The one-link flap is a separate availability issue. Confirm with 'show lag', 'show lacp', 'show mlag', and the configured hash; compare flow counts and member utilization over time before changing the design.
2. What would you verify before adding a third link, and what changes are most likely to improve distribution?
Read the worked answer
Verify the MLAG ID and LACP system ID are consistent on both peers, the server bond is active, and both member ports have identical VLAN, MTU, and STP settings. Check whether the hash uses L2 only or L3 and L4; enabling a stronger L3 or L4 hash can spread the eight flows better if supported and if the traffic is IP. Adding a third 25G member only helps if aggregate demand or the flow mix needs it; it will not guarantee even per-flow use. More flows, or per-flow awareness at the server, may matter more. Use 'show lag' and interface counters to baseline before and after.
3. What are the trade-offs of changing LACP timers to fast?
Read the worked answer
Fast timers detect a dead member in about 3 seconds instead of 90 seconds, which can reduce black-holing during a link failure, but they increase LACP PDU processing and can cause unnecessary churn on marginal optics or long-haul links. In this case the 45-second flap was likely not detected quickly with long timers; fast timers plus a stable link are reasonable. Do not use fast timers as a substitute for fixing a flapping physical link, and test the change in a maintenance window because a timer mismatch can keep a member from bundling.