MC-LAG Uplink Saturation on a Campus Aggregation
A university building aggregates 48 Aruba CX 6200 access switches into a pair of CX 8360 aggregation switches via 2x10G MC-LAG uplinks per access switch. The building also has a 10G fiber ring to the core. During exam week, students report slow LMS video and intermittent VoIP drops. Monitoring shows the MC-LAG uplinks to the aggregation running at 85-95% sustained, with microbursts to 100%. VoIP uses a dedicated voice VLAN with DSCP EF; video uses DSCP AF41. The core link to the building is only 40% utilized. QoS is currently trust-DSCP on access uplinks but there is no egress shaping or queue tuning on the aggregation.
1. Where is the actual bottleneck, and why does the core link being underutilized not help?
Read the worked answer
The bottleneck is the access-to-aggregation MC-LAG uplink (2x10G per access switch), not the core. Even though the building's core fiber is only 40% utilized, traffic must first traverse the oversubscribed MC-LAG to reach the aggregation and then the core.
Reasoning:
- 85-95% sustained with 100% microbursts means the uplink is the constrained segment; the core has headroom but is downstream of the constraint.
- MC-LAG hashing sends flows across both 10G members; a single elephant flow (e.g., a large video stream) can pin to one member and saturate it while the other is idle, producing the microbursts.
- VoIP drops occur because EF traffic shares the congested uplink with best-effort video; without egress queue reservation, EF is dropped in bursts.
Judgement: adding core bandwidth will not fix it. The fix must be at the access uplink (capacity, hashing, or QoS).
2. Propose a prioritized remediation plan that improves VoIP and video without a forklift upgrade, and state how you would validate it.
Read the worked answer
Priority 1 - Protect voice (immediate, low risk):
- On the aggregation egress toward access, configure QoS queues with strict priority for EF (voice) and a guaranteed minimum for AF41 (video).
- Enable egress shaping on the aggregation MC-LAG members to smooth microbursts (shape slightly below line rate, e.g., 9.5G per 10G member).
- Confirm
trust dscpon access uplinks and that the voice VLAN is marked EF end-to-end.
Priority 2 - Reduce uplink pressure:
- Enable L3/L4 hashing on the MC-LAG so video flows spread across members (avoid all-elephant-on-one-member).
- Apply ingress rate-limiting or per-role bandwidth contracts for student video if policy allows.
- Move latency-sensitive video to a locally cached/CDN path if available.
Priority 3 - Capacity (if sustained >80% after above):
- Add a third 10G member per access switch or migrate high-demand access switches to 2x25G.
Validation:
show interfaceutilization andshow qos queueon aggregation - EF queue should show near-zero drops; AF41 within guaranteed share.show lacp aggregatesto confirm member utilization is balanced after hashing change.- Synthetic VoIP MOS test and video stall metrics during peak; confirm microburst counters drop.
- Re-check that core link remains <60% (it should, since we did not add core load).