Cisco 300-640 Practice Test 15
Question 140
The acceptance board must decide whether an AI-infrastructure response to tail latency and network jitter is fit for controlled operation. The fictional operating facts are: A customer-search inference tier reports 8-ms median RPC latency but occasional 250-ms network spikes under microbursts. Signoff cannot rely on a good-looking utilization graph, a change ticket alone or an unmeasured promise. The response must include an appropriate governing engineering principle, a scoped intervention and a separate measurable acceptance criterion. Choose EXACTLY THREE statements that meet those roles.
- Measured p99 inference response remains within budget during the approved peak-burst workload.
- GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS.
- A low-average-latency network can still fail serving SLOs when transient queuing creates high p99 latency.
- Reserve appropriate service classes and relieve burst contention after validating which path contributes to tail delay.
- Apply least-privilege segmentation and test allowed model/data flows plus blocked management and tenant access.
- The rack survives an authorized single-feed test without exceeding the remaining electrical capacity.
Explanation
Correct Answer
A:A customer-search inference tier reports 8-ms median RPC latency but occasional 250-ms network spikes under microbursts. Measured p99 inference response remains within budget during the approved peak-burst workload. Reserve appropriate service classes and relieve burst contention after validating which path contributes to tail delay.
C:A customer-search inference tier reports 8-ms median RPC latency but occasional 250-ms network spikes under microbursts. A low-average-latency network can still fail serving SLOs when transient queuing creates high p99 latency. Reserve appropriate service classes and relieve burst contention after validating which path contributes to tail delay.
D:A customer-search inference tier reports 8-ms median RPC latency but occasional 250-ms network spikes under microbursts. Reserve appropriate service classes and relieve burst contention after validating which path contributes to tail delay. A low-average-latency network can still fail serving SLOs when transient queuing creates high p99 latency.
Incorrect Answers
B:GPU selection must consider model parameters, activation memory, optimizer states, precision and parallelism rather than nominal FLOPS. It is not the rule that supports the proposed change and acceptance test: A customer-search inference tier reports 8-ms median RPC latency but occasional 250-ms network spikes under microbursts. A low-average-latency network can still fail serving SLOs when transient queuing creates high p99 latency. Reserve appropriate service classes and relieve burst contention after validating which path contributes to tail delay.
E:Apply least-privilege segmentation and test allowed model/data flows plus blocked management and tenant access. It does not implement the governing rule for this approval: A customer-search inference tier reports 8-ms median RPC latency but occasional 250-ms network spikes under microbursts. Reserve appropriate service classes and relieve burst contention after validating which path contributes to tail delay. Measured p99 inference response remains within budget during the approved peak-burst workload.
F:The rack survives an authorized single-feed test without exceeding the remaining electrical capacity. It would not independently demonstrate that the control works: A customer-search inference tier reports 8-ms median RPC latency but occasional 250-ms network spikes under microbursts. Measured p99 inference response remains within budget during the approved peak-burst workload. Reserve appropriate service classes and relieve burst contention after validating which path contributes to tail delay.
Question 141
A data-center AI architecture committee considers redundant data-center network paths in its AI Infrastructure Components and Architecture review. At the affected site: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. The committee is deliberately not selecting a vendor product or applying a live change yet. It needs the foundational reasoning that makes the next design decision defensible. Which ONE engineering principle directly governs this situation? Choose ONE answer.
- A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence.
- Distributed AI placement must limit correlated failures and balance locality against rack, host and power dependencies.
- High file-open rates can bottleneck metadata services even when aggregate data bandwidth is low.
- AI jobs split across sites need an explicit source-of-truth version and replication consistency contract.
Explanation
Correct Answer
A:A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence. Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements.
Incorrect Answers
B:Distributed AI placement must limit correlated failures and balance locality against rack, host and power dependencies. It does not govern the condition described: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence.
C:High file-open rates can bottleneck metadata services even when aggregate data bandwidth is low. It does not govern the condition described: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence.
D:AI jobs split across sites need an explicit source-of-truth version and replication consistency contract. It does not govern the condition described: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence.
Question 142
An infrastructure diagnostician is investigating redundant data-center network paths for the AI Infrastructure Components and Architecture team. The reported incident reads: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. Competing teams attribute the symptoms to different layers of the AI stack. Before accepting a hypothesis or touching production settings, the investigator must choose a discriminating set of measurements. Which ONE evidence package is MOST relevant? Choose ONE answer.
- Observe GPU idle gaps, CPU saturation, page faults, dataloader queue depth, NUMA affinity and PCIe locality.
- Correlate request p99 with per-hop queue occupancy, drops, ECN marks and burst timing rather than average link load.
- Collect calibrated facility and IT power readings, interval timestamps, seasonal conditions and utilization context.
- Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements.
Explanation
Correct Answer
D:A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements. A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence.
Incorrect Answers
A:Observe GPU idle gaps, CPU saturation, page faults, dataloader queue depth, NUMA affinity and PCIe locality. That check would not establish the decisive facts for this condition: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements.
B:Correlate request p99 with per-hop queue occupancy, drops, ECN marks and burst timing rather than average link load. That check would not establish the decisive facts for this condition: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements.
C:Collect calibrated facility and IT power readings, interval timestamps, seasonal conditions and utilization context. That check would not establish the decisive facts for this condition: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements.
Question 143
During a bounded AI-platform change-control review concerning redundant data-center network paths, the site owner reports: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. The proposed work must have a narrow implementation scope, first be validated against the failure evidence and avoid unrelated system-wide tuning. Which ONE intervention should the engineers pilot before any wider rollout? Choose ONE answer.
- Compare like-for-like PUE baselines and optimize cooling or capacity with a documented energy measurement plan.
- Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
- Set class-based retention and quotas while keeping legally required provenance and restorable checkpoints.
- Tune CPU worker allocation, memory pinning and NUMA or device placement before buying additional GPUs.
Explanation
Correct Answer
B:A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic. Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements.
Incorrect Answers
A:Compare like-for-like PUE baselines and optimize cooling or capacity with a documented energy measurement plan. That change would leave the reported problem unresolved: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
C:Set class-based retention and quotas while keeping legally required provenance and restorable checkpoints. That change would leave the reported problem unresolved: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
D:Tune CPU worker allocation, memory pinning and NUMA or device placement before buying additional GPUs. That change would leave the reported problem unresolved: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
Question 144
Two independent technical reviewers examine a AI Infrastructure Components and Architecture problem involving redundant data-center network paths. Their shared incident log states: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. One reviewer requires measurements capable of separating the observed problem from neighboring failure modes; the other requires a bounded action that follows those measurements. Choose EXACTLY TWO statements that together supply those two parts.
- Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision.
- Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
- Apply least-privilege segmentation and test allowed model/data flows plus blocked management and tenant access.
- Review rack nameplate ratings, measured peak draw, dual-feed loading, breaker headroom and PSU redundancy.
- Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements.
Explanation
Correct Answer
B:A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic. Training or serving services survive an approved single-fault test within the documented recovery objective.
E:A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements. Training or serving services survive an approved single-fault test within the documented recovery objective.
Incorrect Answers
A:Collect out-of-memory traces, per-rank allocator usage, peak activations, tensor sizes and chosen precision. It would not provide the evidence needed for the paired diagnosis and follow-up: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements. Training or serving services survive an approved single-fault test within the documented recovery objective.
C:Apply least-privilege segmentation and test allowed model/data flows plus blocked management and tenant access. It is not the scoped response supported by the accompanying evidence: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic. Training or serving services survive an approved single-fault test within the documented recovery objective.
D:Review rack nameplate ratings, measured peak draw, dual-feed loading, breaker headroom and PSU redundancy. It would not provide the evidence needed for the paired diagnosis and follow-up: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements. Training or serving services survive an approved single-fault test within the documented recovery objective.
Question 145
The acceptance board must decide whether an AI-infrastructure response to redundant data-center network paths is fit for controlled operation. The fictional operating facts are: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. Signoff cannot rely on a good-looking utilization graph, a change ticket alone or an unmeasured promise. The response must include an appropriate governing engineering principle, a scoped intervention and a separate measurable acceptance criterion. Choose EXACTLY THREE statements that meet those roles.
- A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence.
- Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
- Multiple tenants operate within their allocations while a full-device benchmark remains unaffected.
- Use an approved durable checkpoint repository with atomic or validated writes and restore tests.
- AI thermal designs must handle rack heat flux and liquid or air-cooling capacity under worst-case workloads.
- Training or serving services survive an approved single-fault test within the documented recovery objective.
Explanation
Correct Answer
A:A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
B:A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic. A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence.
F:A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. Training or serving services survive an approved single-fault test within the documented recovery objective. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
Incorrect Answers
C:Multiple tenants operate within their allocations while a full-device benchmark remains unaffected. It would not independently demonstrate that the control works: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. Training or serving services survive an approved single-fault test within the documented recovery objective. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
D:Use an approved durable checkpoint repository with atomic or validated writes and restore tests. It does not implement the governing rule for this approval: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic. Training or serving services survive an approved single-fault test within the documented recovery objective.
E:AI thermal designs must handle rack heat flux and liquid or air-cooling capacity under worst-case workloads. It is not the rule that supports the proposed change and acceptance test: A financial AI cluster has two uplinks but both traverse the same maintenance switch and lose reachability together. A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
Question 146
A data-center AI architecture committee considers redundant data-center network paths in its AI Infrastructure Components and Architecture review. At the affected site: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. The committee is deliberately not selecting a vendor product or applying a live change yet. It needs the foundational reasoning that makes the next design decision defensible. Which ONE engineering principle directly governs this situation? Choose ONE answer.
- AI servers, firmware, NIC drivers, GPU stack and fabric software form a versioned support matrix.
- Training storage must sustain aggregate sequential and random reads across workers, not just advertised peak bandwidth.
- A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence.
- Checkpoint systems must meet recovery-point and recovery-time needs with verified durable writes and restore validation.
Explanation
Correct Answer
C:A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence. Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements.
Incorrect Answers
A:AI servers, firmware, NIC drivers, GPU stack and fabric software form a versioned support matrix. It does not govern the condition described: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence.
B:Training storage must sustain aggregate sequential and random reads across workers, not just advertised peak bandwidth. It does not govern the condition described: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence.
D:Checkpoint systems must meet recovery-point and recovery-time needs with verified durable writes and restore validation. It does not govern the condition described: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence.
Question 147
An infrastructure diagnostician is investigating redundant data-center network paths for the AI Infrastructure Components and Architecture team. The reported incident reads: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. Competing teams attribute the symptoms to different layers of the AI stack. Before accepting a hypothesis or touching production settings, the investigator must choose a discriminating set of measurements. Which ONE evidence package is MOST relevant? Choose ONE answer.
- Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements.
- Collect calibrated facility and IT power readings, interval timestamps, seasonal conditions and utilization context.
- Measure per-worker read concurrency, storage-node load, cache hit rates, queue depth and samples-per-second.
- Review rack nameplate ratings, measured peak draw, dual-feed loading, breaker headroom and PSU redundancy.
Explanation
Correct Answer
A:A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements. A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence.
Incorrect Answers
B:Collect calibrated facility and IT power readings, interval timestamps, seasonal conditions and utilization context. That check would not establish the decisive facts for this condition: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements.
C:Measure per-worker read concurrency, storage-node load, cache hit rates, queue depth and samples-per-second. That check would not establish the decisive facts for this condition: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements.
D:Review rack nameplate ratings, measured peak draw, dual-feed loading, breaker headroom and PSU redundancy. That check would not establish the decisive facts for this condition: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements.
Question 148
During a bounded AI-platform change-control review concerning redundant data-center network paths, the site owner reports: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. The proposed work must have a narrow implementation scope, first be validated against the failure evidence and avoid unrelated system-wide tuning. Which ONE intervention should the engineers pilot before any wider rollout? Choose ONE answer.
- Package or shard files appropriately and scale the supported metadata path while preserving data semantics.
- Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
- Validate and stage a supported end-to-end software bill of materials with rollback images.
- Benchmark representative dataset shards and tune supported striping, caching or storage tiers based on bottlenecks.
Explanation
Correct Answer
B:A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic. Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements.
Incorrect Answers
A:Package or shard files appropriately and scale the supported metadata path while preserving data semantics. That change would leave the reported problem unresolved: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
C:Validate and stage a supported end-to-end software bill of materials with rollback images. That change would leave the reported problem unresolved: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
D:Benchmark representative dataset shards and tune supported striping, caching or storage tiers based on bottlenecks. That change would leave the reported problem unresolved: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
Question 149
Two independent technical reviewers examine a AI Infrastructure Components and Architecture problem involving redundant data-center network paths. Their shared incident log states: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. One reviewer requires measurements capable of separating the observed problem from neighboring failure modes; the other requires a bounded action that follows those measurements. Choose EXACTLY TWO statements that together supply those two parts.
- Check checkpoint frequency, write completion, replicas, integrity hashes and actual full-state restore time.
- Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements.
- Measure per-worker read concurrency, storage-node load, cache hit rates, queue depth and samples-per-second.
- Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
- Package or shard files appropriately and scale the supported metadata path while preserving data semantics.
Explanation
Correct Answer
B:A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements. Training or serving services survive an approved single-fault test within the documented recovery objective.
D:A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic. Training or serving services survive an approved single-fault test within the documented recovery objective.
Incorrect Answers
A:Check checkpoint frequency, write completion, replicas, integrity hashes and actual full-state restore time. It would not provide the evidence needed for the paired diagnosis and follow-up: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements. Training or serving services survive an approved single-fault test within the documented recovery objective.
C:Measure per-worker read concurrency, storage-node load, cache hit rates, queue depth and samples-per-second. It would not provide the evidence needed for the paired diagnosis and follow-up: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. Inspect topology failure domains, ECMP next hops, routing timers and failover packet-loss measurements. Training or serving services survive an approved single-fault test within the documented recovery objective.
E:Package or shard files appropriately and scale the supported metadata path while preserving data semantics. It is not the scoped response supported by the accompanying evidence: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic. Training or serving services survive an approved single-fault test within the documented recovery objective.
Question 150
The acceptance board must decide whether an AI-infrastructure response to redundant data-center network paths is fit for controlled operation. The fictional operating facts are: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. Signoff cannot rely on a good-looking utilization graph, a change ticket alone or an unmeasured promise. The response must include an appropriate governing engineering principle, a scoped intervention and a separate measurable acceptance criterion. Choose EXACTLY THREE statements that meet those roles.
- Measured p99 inference response remains within budget during the approved peak-burst workload.
- Reserve and prewarm sufficient inference replicas across independent hosts to meet stated SLO during failover.
- A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence.
- Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
- Training or serving services survive an approved single-fault test within the documented recovery objective.
- Distributed AI placement must limit correlated failures and balance locality against rack, host and power dependencies.
Explanation
Correct Answer
C:A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
D:A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic. A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence.
E:A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. Training or serving services survive an approved single-fault test within the documented recovery objective. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
Incorrect Answers
A:Measured p99 inference response remains within budget during the approved peak-burst workload. It would not independently demonstrate that the control works: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. Training or serving services survive an approved single-fault test within the documented recovery objective. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.
B:Reserve and prewarm sufficient inference replicas across independent hosts to meet stated SLO during failover. It does not implement the governing rule for this approval: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic. Training or serving services survive an approved single-fault test within the documented recovery objective.
F:Distributed AI placement must limit correlated failures and balance locality against rack, host and power dependencies. It is not the rule that supports the proposed change and acceptance test: A genomics facility labels its GPU pods dual-homed although both fabric connections rely on one shared power feed. A cluster should tolerate a single link or fabric-device failure using validated redundant paths and routing convergence. Build physically independent eligible paths and test controlled link or leaf failure under active AI traffic.