ProDeploy — Learn the fabric that keeps AI clusters alive

← Labs ← Chapter 15: IP Routing for AI/ML FabricsBGP suboptimal routing: spine ASN design

Desktop recommendedChecking session

BGP suboptimal routing: spine ASN design

advanced

Elapsed

00:00

Scenario

It is 02:15. An LLM training job on a 256-GPU cluster is experiencing 40% throughput degradation and intermittent gradient synchronisation timeouts. The job started 6 hours ago and is at 78% completion. Stopping it would waste ~18 hours of compute time. A monitoring alert fired 20 minutes ago: 'SpineB -> Leaf2 link utilisation: 89%'. Your task: determine why SpineB is overloaded despite no server-side changes, find the root cause in the BGP routing design, and implement the correct fix. This lab intentionally includes a non-standard SpineA-SpineB eBGP peering so you can observe how mismatched spine ASNs let a failed-spine detour route survive.

Conditions

Failure scope checked
Suboptimal path identified
Same-ASN spine principle identified
Spine A ASN unified
Spine B ASN unified
Post-fix path verification complete

Need Help?

Open the step-by-step command guide for this lab.

Topology

Filled green circle = Active

Unfilled red circle = Down

Solid line = 400G link

Dashed line = degraded

Click a device node to open its terminal session

Preparing lab environment...