A school-district lab reseats a suspect InfiniBand cable and needs proof the path improved. What measurement should they compare?
Select an answer to reveal the explanation.
Short Explanation
Cable reseats are easy to brag about and hard to prove. Run ib_write_lat before and after so the school-district lab has hard InfiniBand write-latency numbers, not vibes.
Full Explanation
After a school-district lab reseats a suspect InfiniBand cable, proof of path improvement must come from InfiniBand performance measurement, not unrelated telemetry. ib_write_lat measures InfiniBand write latency between endpoints and yields comparable before/after numbers that show whether the physical-path remediation reduced latency or stabilized the RDMA path used by lab workloads. That directly tests the remediated InfiniBand segment. ib_write_bw throughput averages collected after the cable reseat capture only a single post-change throughput snapshot and lack the before/after latency comparison needed to show the InfiniBand path improved. Ethernet ping round-trip times between the cluster management ports test a different network and different endpoints, so they do not validate the remediated InfiniBand cable path. perfquery link error counters sampled from an adjacent idle switch port measure a different, idle port rather than the suspect cable endpoints and cannot prove the IB path improved. Exam caveat: keep endpoints, message sizes, and test duration consistent across before/after runs; otherwise latency deltas are not attributable to the cable reseat. Operational check: capture baseline ib_write_lat results, reseat the InfiniBand cable, re-run the same latency test between the same school-district nodes, and document whether latency and error indicators improved before closing the ticket.