Cisco 300-640 Practice Test 32
Question 310
The acceptance board must decide whether an AI-infrastructure response to priority flow control scope is fit for controlled operation. The fictional operating facts are: A new RoCEv2 training rack has sporadic loss on its selected priority, and operators propose global link-level pausing instead of inspecting class mapping. 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.
- Nodes remain within the documented clock skew tolerance and cross-system events align.
- Align endpoint and fabric settings on a documented RoCEv2 path before high-rate collective traffic.
- Server boot and storage policies must match approved boot targets and multipath recovery requirements.
- RoCE deployments may need priority flow control on designated loss-sensitive classes, but PFC is not a blanket cure for congestion.
- RoCE traffic meets loss and job-completion targets while pause propagation stays bounded.
- Scope PFC only to validated loss-sensitive priorities and tune headroom while preventing congestion spreading to unrelated flows.
Explanation
Correct Answer
D:A new RoCEv2 training rack has sporadic loss on its selected priority, and operators propose global link-level pausing instead of inspecting class mapping. RoCE deployments may need priority flow control on designated loss-sensitive classes, but PFC is not a blanket cure for congestion. Scope PFC only to validated loss-sensitive priorities and tune headroom while preventing congestion spreading to unrelated flows.
E:A new RoCEv2 training rack has sporadic loss on its selected priority, and operators propose global link-level pausing instead of inspecting class mapping. RoCE traffic meets loss and job-completion targets while pause propagation stays bounded. Scope PFC only to validated loss-sensitive priorities and tune headroom while preventing congestion spreading to unrelated flows.
F:A new RoCEv2 training rack has sporadic loss on its selected priority, and operators propose global link-level pausing instead of inspecting class mapping. Scope PFC only to validated loss-sensitive priorities and tune headroom while preventing congestion spreading to unrelated flows. RoCE deployments may need priority flow control on designated loss-sensitive classes, but PFC is not a blanket cure for congestion.
Incorrect Answers
A:Nodes remain within the documented clock skew tolerance and cross-system events align. It would not independently demonstrate that the control works: A new RoCEv2 training rack has sporadic loss on its selected priority, and operators propose global link-level pausing instead of inspecting class mapping. RoCE traffic meets loss and job-completion targets while pause propagation stays bounded. Scope PFC only to validated loss-sensitive priorities and tune headroom while preventing congestion spreading to unrelated flows.
B:Align endpoint and fabric settings on a documented RoCEv2 path before high-rate collective traffic. It does not implement the governing rule for this approval: A new RoCEv2 training rack has sporadic loss on its selected priority, and operators propose global link-level pausing instead of inspecting class mapping. Scope PFC only to validated loss-sensitive priorities and tune headroom while preventing congestion spreading to unrelated flows. RoCE traffic meets loss and job-completion targets while pause propagation stays bounded.
C:Server boot and storage policies must match approved boot targets and multipath recovery requirements. It is not the rule that supports the proposed change and acceptance test: A new RoCEv2 training rack has sporadic loss on its selected priority, and operators propose global link-level pausing instead of inspecting class mapping. RoCE deployments may need priority flow control on designated loss-sensitive classes, but PFC is not a blanket cure for congestion. Scope PFC only to validated loss-sensitive priorities and tune headroom while preventing congestion spreading to unrelated flows.
Question 311
A data-center AI architecture committee considers explicit congestion notification tuning in its AI Infrastructure Deployment and Data Management review. At the affected site: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. 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.
- UCS system-class and vNIC QoS settings must be coherent with switch-side traffic treatment for AI and storage flows.
- Changes to loss-sensitive AI traffic need dependency mapping, rollback and independent acceptance measurements.
- ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic.
- Server boot and storage policies must match approved boot targets and multipath recovery requirements.
Explanation
Correct Answer
C:A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic. Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts.
Incorrect Answers
A:UCS system-class and vNIC QoS settings must be coherent with switch-side traffic treatment for AI and storage flows. It does not govern the condition described: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic.
B:Changes to loss-sensitive AI traffic need dependency mapping, rollback and independent acceptance measurements. It does not govern the condition described: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic.
D:Server boot and storage policies must match approved boot targets and multipath recovery requirements. It does not govern the condition described: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic.
Question 312
An infrastructure diagnostician is investigating explicit congestion notification tuning for the AI Infrastructure Deployment and Data Management team. The reported incident reads: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. 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.
- Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts.
- Review claimed server identity, organization roles, assigned policies, compliance drift and rollout scope.
- Check UCS domain profiles, fabric interconnect state, policy inheritance, version support and planned change impact.
- Compare requested topology, supported switch and server models, firmware, connectivity, identity and change ownership.
Explanation
Correct Answer
A:A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts. ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic.
Incorrect Answers
B:Review claimed server identity, organization roles, assigned policies, compliance drift and rollout scope. That check would not establish the decisive facts for this condition: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts.
C:Check UCS domain profiles, fabric interconnect state, policy inheritance, version support and planned change impact. That check would not establish the decisive facts for this condition: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts.
D:Compare requested topology, supported switch and server models, firmware, connectivity, identity and change ownership. That check would not establish the decisive facts for this condition: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts.
Question 313
During a bounded AI-platform change-control review concerning explicit congestion notification tuning, the site owner reports: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. 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.
- Align supported UCS QoS classes with end-to-end data-center QoS policy and test contention.
- Stage a validated domain profile update with diff review and a reversible maintenance workflow.
- Pilot one representative rack, measure controlled training traffic and expand only after validated acceptance.
- Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.
Explanation
Correct Answer
D:A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load. Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts.
Incorrect Answers
A:Align supported UCS QoS classes with end-to-end data-center QoS policy and test contention. That change would leave the reported problem unresolved: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.
B:Stage a validated domain profile update with diff review and a reversible maintenance workflow. That change would leave the reported problem unresolved: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.
C:Pilot one representative rack, measure controlled training traffic and expand only after validated acceptance. That change would leave the reported problem unresolved: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.
Question 314
Two independent technical reviewers examine a AI Infrastructure Deployment and Data Management problem involving explicit congestion notification tuning. Their shared incident log states: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. 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.
- Verify onboarded site inventory, supported features, access roles, software versions and configuration diffs.
- Choose supported load-distribution behavior and adjust flow placement or links without creating packet reordering hazards.
- Inspect class mapping, offered traffic rate, bandwidth allocation and starvation on storage and control classes.
- Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.
- Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts.
Explanation
Correct Answer
D:A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load. Queue depth and tail latency decline without unacceptable throughput loss or persistent marks.
E:A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts. Queue depth and tail latency decline without unacceptable throughput loss or persistent marks.
Incorrect Answers
A:Verify onboarded site inventory, supported features, access roles, software versions and configuration diffs. It would not provide the evidence needed for the paired diagnosis and follow-up: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts. Queue depth and tail latency decline without unacceptable throughput loss or persistent marks.
B:Choose supported load-distribution behavior and adjust flow placement or links without creating packet reordering hazards. It is not the scoped response supported by the accompanying evidence: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load. Queue depth and tail latency decline without unacceptable throughput loss or persistent marks.
C:Inspect class mapping, offered traffic rate, bandwidth allocation and starvation on storage and control classes. It would not provide the evidence needed for the paired diagnosis and follow-up: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts. Queue depth and tail latency decline without unacceptable throughput loss or persistent marks.
Question 315
The acceptance board must decide whether an AI-infrastructure response to explicit congestion notification tuning is fit for controlled operation. The fictional operating facts are: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. 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.
- Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.
- Use approved redundant time sources and consistent UCS or host NTP policy, then verify convergence.
- Cross-subnet RDMA tests meet loss and latency requirements without fragmentation or class mismatch.
- Queue depth and tail latency decline without unacceptable throughput loss or persistent marks.
- ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic.
- Changes to loss-sensitive AI traffic need dependency mapping, rollback and independent acceptance measurements.
Explanation
Correct Answer
A:A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load. ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic.
D:A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. Queue depth and tail latency decline without unacceptable throughput loss or persistent marks. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.
E:A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.
Incorrect Answers
B:Use approved redundant time sources and consistent UCS or host NTP policy, then verify convergence. It does not implement the governing rule for this approval: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load. Queue depth and tail latency decline without unacceptable throughput loss or persistent marks.
C:Cross-subnet RDMA tests meet loss and latency requirements without fragmentation or class mismatch. It would not independently demonstrate that the control works: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. Queue depth and tail latency decline without unacceptable throughput loss or persistent marks. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.
F:Changes to loss-sensitive AI traffic need dependency mapping, rollback and independent acceptance measurements. It is not the rule that supports the proposed change and acceptance test: A gradient synchronization workload sees bursts of drops after queues grow, although compatible NIC endpoints support ECN feedback. ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.
Question 316
A data-center AI architecture committee considers explicit congestion notification tuning in its AI Infrastructure Deployment and Data Management review. At the affected site: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. 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.
- ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic.
- UCS power policy must reflect chassis limits, redundant feed capacity and peak AI workload characteristics.
- Time synchronization across UCS, fabric, GPU hosts and monitoring is needed for reliable correlation and operational security.
- An AI fabric must classify and preserve traffic intent end to end rather than trust inconsistent markings from every source.
Explanation
Correct Answer
A:An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic. Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts.
Incorrect Answers
B:UCS power policy must reflect chassis limits, redundant feed capacity and peak AI workload characteristics. It does not govern the condition described: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic.
C:Time synchronization across UCS, fabric, GPU hosts and monitoring is needed for reliable correlation and operational security. It does not govern the condition described: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic.
D:An AI fabric must classify and preserve traffic intent end to end rather than trust inconsistent markings from every source. It does not govern the condition described: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic.
Question 317
An infrastructure diagnostician is investigating explicit congestion notification tuning for the AI Infrastructure Deployment and Data Management team. The reported incident reads: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. 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 vNIC-to-fabric mappings, adapter queue settings, MAC identity, VLAN membership and actual path MTU.
- Inspect class mapping, offered traffic rate, bandwidth allocation and starvation on storage and control classes.
- Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts.
- Verify source and destination NIC RoCE mode, IP reachability, path MTU, DSCP/priority mapping and observed retransmission or loss.
Explanation
Correct Answer
C:An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts. ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic.
Incorrect Answers
A:Inspect vNIC-to-fabric mappings, adapter queue settings, MAC identity, VLAN membership and actual path MTU. That check would not establish the decisive facts for this condition: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts.
B:Inspect class mapping, offered traffic rate, bandwidth allocation and starvation on storage and control classes. That check would not establish the decisive facts for this condition: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts.
D:Verify source and destination NIC RoCE mode, IP reachability, path MTU, DSCP/priority mapping and observed retransmission or loss. That check would not establish the decisive facts for this condition: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts.
Question 318
During a bounded AI-platform change-control review concerning explicit congestion notification tuning, the site owner reports: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. 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.
- Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.
- Apply a supported UCS power policy after confirming rack and chassis electrical budgets.
- Configure supported ETS class weights to protect essential shares without starving management and storage traffic.
- Define trusted classification boundaries and validate consistent class-to-queue behavior across endpoints and switches.
Explanation
Correct Answer
A:An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load. Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts.
Incorrect Answers
B:Apply a supported UCS power policy after confirming rack and chassis electrical budgets. That change would leave the reported problem unresolved: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.
C:Configure supported ETS class weights to protect essential shares without starving management and storage traffic. That change would leave the reported problem unresolved: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.
D:Define trusted classification boundaries and validate consistent class-to-queue behavior across endpoints and switches. That change would leave the reported problem unresolved: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.
Question 319
Two independent technical reviewers examine a AI Infrastructure Deployment and Data Management problem involving explicit congestion notification tuning. Their shared incident log states: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. 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.
- Review power-cap settings, power telemetry, thermal alarms and PSU redundancy against expected GPU draw.
- Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts.
- Inspect vNIC-to-fabric mappings, adapter queue settings, MAC identity, VLAN membership and actual path MTU.
- Align endpoint and fabric settings on a documented RoCEv2 path before high-rate collective traffic.
- Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.
Explanation
Correct Answer
B:An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts. Queue depth and tail latency decline without unacceptable throughput loss or persistent marks.
E:An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load. Queue depth and tail latency decline without unacceptable throughput loss or persistent marks.
Incorrect Answers
A:Review power-cap settings, power telemetry, thermal alarms and PSU redundancy against expected GPU draw. It would not provide the evidence needed for the paired diagnosis and follow-up: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts. Queue depth and tail latency decline without unacceptable throughput loss or persistent marks.
C:Inspect vNIC-to-fabric mappings, adapter queue settings, MAC identity, VLAN membership and actual path MTU. It would not provide the evidence needed for the paired diagnosis and follow-up: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. Correlate ECN marking, switch queue depth, RTT, endpoint congestion control and packet loss during collective bursts. Queue depth and tail latency decline without unacceptable throughput loss or persistent marks.
D:Align endpoint and fabric settings on a documented RoCEv2 path before high-rate collective traffic. It is not the scoped response supported by the accompanying evidence: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load. Queue depth and tail latency decline without unacceptable throughput loss or persistent marks.
Question 320
The acceptance board must decide whether an AI-infrastructure response to explicit congestion notification tuning is fit for controlled operation. The fictional operating facts are: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. 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.
- Pilot one representative rack, measure controlled training traffic and expand only after validated acceptance.
- Maximum approved training load stays within electrical limits without involuntary power capping or host resets.
- RoCE deployments may need priority flow control on designated loss-sensitive classes, but PFC is not a blanket cure for congestion.
- Queue depth and tail latency decline without unacceptable throughput loss or persistent marks.
- Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.
- ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic.
Explanation
Correct Answer
D:An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. Queue depth and tail latency decline without unacceptable throughput loss or persistent marks. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.
E:An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load. ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic.
F:An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.
Incorrect Answers
A:Pilot one representative rack, measure controlled training traffic and expand only after validated acceptance. It does not implement the governing rule for this approval: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load. Queue depth and tail latency decline without unacceptable throughput loss or persistent marks.
B:Maximum approved training load stays within electrical limits without involuntary power capping or host resets. It would not independently demonstrate that the control works: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. Queue depth and tail latency decline without unacceptable throughput loss or persistent marks. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.
C:RoCE deployments may need priority flow control on designated loss-sensitive classes, but PFC is not a blanket cure for congestion. It is not the rule that supports the proposed change and acceptance test: An AI fabric marks almost all packets under light load and collective throughput drops despite no sustained congested link. ECN should signal incipient congestion so capable endpoints react before queues overflow, with marking thresholds validated against workload traffic. Apply version-supported ECN thresholds for the designated traffic class and validate endpoint response under controlled load.