Overview

  • We have reproduced the issue in more than 3 instances (multiple times in each instance).
    • Flaps were only seen only with Credo AEC cables
    • We had DAC cables and DR8 Optics in the same setup, which did not observe these flaps.
  • The behavior is the same: local_reason_opcode = alignment loss on the NIC end (Amber log).
    • NIC initializes; link goes down.
    • Link comes back up.
    • Resulting in a ~12-second link flap.
  • One leg of the AEC cable was connected to BF-3; the other was dangling.
  • Voltage margin variation: -5% to +3.16% of nominal.
  • Built a scaled-up FA unit (2 occurrences of flap on one AEC, same signature as above) with:
    • 32x 800G AEC to 2x400G BF-3140
    • 22x 800G AEC to QFX5240-64QD
    • 28x 800G OSFP 800G DR8 to QFX5240-64OD

XAI Topology

XAI Topology