A city GPU InfiniBand fabric needs recurring civic ops visibility into system health. What should operators schedule?
Select an answer to reveal the explanation.
Short Explanation
Civic ops likes a daily fabric pulse check. Schedule daily UFM system health reviews so the city GPU InfiniBand fabric stays visible to municipal operations.
Full Explanation
A city GPU InfiniBand fabric needs recurring civic ops visibility into system health. Scheduling daily UFM system health reviews for InfiniBand municipal fabric operations catches SM/HA drift, degraded services, and emerging instability before citizen-facing or departmental AI jobs fail in production and before issues become multi-hour outages.
Why the other options fail by concept: relying on weekly UFM link health summaries for municipal InfiniBand fabric planning fails because weekly summaries are planning aids, not recurring system-health reviews; they miss SM/HA drift, firmware aging, and degraded services between looks, leaving citizen-facing jobs exposed. Relying on monthly UFM congestion trend reports for municipal InfiniBand fabric tuning fails because congestion trends support performance tuning, not system-health visibility; they do not verify SM/HA or other control-plane service health, and a monthly cadence is too infrequent and too error-prone to catch emerging instability early. Relying on daily UFM topology snapshot backups for municipal InfiniBand change control fails because backups preserve configuration state rather than assess service health; topology snapshots do not check SM/HA, degraded services, or emerging instability, so civic ops has no way to notice degradation until jobs start failing.
Exam caveat: daily review cadence should be risk-based—higher for production city AI fabrics than for idle lab fabrics—but the exam emphasis is recurring UFM system health visibility, not a single or infrequent look. Pair reviews with alerting so nights and weekends are not blind between formal reviews.
Operational check: add a daily UFM system health check to city GPU IB ops, record pass/fail with component notes, trend anomalies week over week, and escalate when health degrades even if jobs have not yet failed so civic ops stays ahead of outages.