NCM-MCI 6.10: Nutanix Certified Master - Multicloud Infrastructure Practice Exam

What the NCM-MCI 6.10 exam covers

• Advanced Cluster Management and Design89 questions
• Advanced AHV and VM Management88 questions
• Advanced Networking92 questions
• Advanced Storage and Performance74 questions
• Advanced Data Protection and Disaster Recovery83 questions
• Security and Compliance74 questions
• Advanced Monitoring and Troubleshooting85 questions
• Lifecycle, Upgrades, and Migration86 questions

Free NCM-MCI 6.10 sample questions

A sample of 10 questions with answers and explanations. Sign up free to practice all 671.

1. Question 1Advanced Cluster Management and Design

   A customer runs a 4-node RF2 cluster where each node holds 8 TB of physical SSD capacity (32 TB raw, 16 TB usable after RF2). They have provisioned guest VMs consuming 13 TB of effective data and are pushing close to 85% utilization. During a planned firmware upgrade one node is taken offline and the cluster begins to alert that it cannot maintain its resiliency target. What is the MOST accurate root cause?

   • AThe cluster did not reserve enough rebuild capacity for a single-node failure, so with one node down the remaining nodes cannot host a second full RF2 copy of all dataCorrect
   • BCurator stopped its background scans once utilization crossed 85%, so the stale replica map prevented the cluster from re-establishing its second copy on the surviving three nodes
   • CRF2 requires a minimum of five nodes to satisfy resiliency, so any 4-node cluster is unable to keep its target intact while a node is offline for firmware
   • DStargate switched every guest write to synchronous oplog draining during the upgrade, which doubled the space each write consumed and exhausted the free capacity
   ✓ Correct answer: A

   Resilient capacity is the amount of usable space a cluster can fill while still being able to re-protect every extent back to its configured RF after losing a fault domain. On a 4-node RF2 cluster you must keep roughly one node's worth of usable space free so Curator can recreate the second replica of the downed node's extents onto the three survivors. At 85% utilization the working set already exceeds that resilient-capacity threshold, so when firmware takes a node offline there is nowhere to rebuild the missing copies and Prism correctly raises a resiliency-target alert. The fix is to lower utilization below resilient capacity or add a node, not to change any background service behavior.

   Why the other options are wrong

   • BCurator continues running background scans well past 85% utilization and never disables them at a fixed threshold, so a stalled scan is not what breaks the resiliency target here.
   • CRF2 has no five-node minimum and a 4-node cluster is a fully supported RF2 configuration, so node count is not the reason it cannot maintain resiliency with one node down.
   • DStargate does not switch to a synchronous oplog draining mode that doubles per-write space during an upgrade; oplog draining to the extent store is asynchronous and space-neutral.
2. Question 2Advanced AHV and VM Management

   A customer configures a VM-VM anti-affinity policy for three database VMs in a five-node cluster so each runs on a separate host. One host fails. After HA restarts the VMs, all three are temporarily running on only two hosts, with two of them co-located. What does ADS do regarding the anti-affinity violation once the failed host is recovered and the cluster has spare capacity?

   • AADS treats VM-VM anti-affinity as a soft (preferential) rule and will migrate one of the co-located VMs to re-establish separation once capacity allowsCorrect
   • BADS leaves the VMs co-located permanently because anti-affinity is only enforced at initial power-on and never re-evaluated
   • CADS powers off one of the co-located VMs until a dedicated host is free to honor the policy
   • DADS converts the anti-affinity rule to a hard rule after an HA event, preventing the VMs from ever sharing a host even during the outage
   ✓ Correct answer: A

   VM-VM anti-affinity in AHV is a soft, preferential rule, which is what lets HA restart all three database VMs even when a host failure temporarily forces two of them to share a host. Once the failed host recovers and spare capacity returns, ADS re-evaluates the policy and migrates one co-located VM to restore the desired separation. The rule is honored as soon as it is feasible rather than blocking restart during the outage. This balances availability during failures with the preferred spread afterward.

   Why the other options are wrong

   • BVM-VM anti-affinity is re-evaluated continuously, not only at initial power-on, so ADS does not leave the VMs co-located permanently once capacity allows separation.
   • CADS does not power off a VM to honor anti-affinity; the rule is soft, so it keeps the VMs running and re-separates them through migration when possible.
   • DAnti-affinity is not promoted to a hard rule after an HA event; if it were hard, HA could not have restarted all three VMs onto two hosts during the outage.
3. Question 3Advanced Networking

   An architect is designing the virtual switch layout for a cluster that must isolate management/CVM traffic, VM guest traffic, and a dedicated backup network onto separate physical uplink pairs. Which design approach is the recommended Nutanix practice on AHV?

   • ACreate additional virtual switches (e.g., vs1, vs2), each assigned its own dedicated uplink ports and bond, separate from vs0Correct
   • BPlace all traffic on vs0 and rely on Flow microsegmentation to separate the networks
   • CUse a single uplink per node and separate traffic purely by VLAN tagging on that one link
   • DDisable vs0 entirely and route all traffic through the CVM's internal 192.168.5.0/24 bridge
   ✓ Correct answer: A

   The recommended AHV practice for physically isolating distinct traffic types onto separate uplink pairs is to create additional virtual switches, each with its own dedicated NICs and bond. Management/CVM traffic stays on vs0, while guest VM and backup traffic move to vs1 and vs2 with their own uplinks, giving true physical separation and independent bandwidth. This matches the requirement to isolate the networks onto separate uplink pairs. It is the supported, scalable way to lay out multiple traffic domains.

   Why the other options are wrong

   • BPlacing all traffic on vs0 and using Flow only segments at the policy layer, not physically; it does not put management, guest, and backup traffic on separate uplink pairs as required.
   • CA single uplink per node with VLAN tagging provides no physical isolation or redundancy, contradicting the requirement for separate physical uplink pairs.
   • DThe 192.168.5.0/24 bridge is the internal host-to-CVM path and cannot carry external management, guest, or backup traffic, so disabling vs0 to use it is not viable.
4. Question 4Advanced Storage and PerformanceSelect all that apply

   A performance engineer is comparing the read path for two scenarios on the same node: (1) data resident in the in-memory unified cache, and (2) data resident only on the local SSD extent store. Which TWO statements correctly describe how Stargate handles these reads? (Choose TWO)

   • AA unified cache hit is served from RAM and avoids touching the extent store, providing the lowest possible read latencyCorrect
   • BA cache miss is read from the local SSD extent store, and the retrieved data is inserted into the unified cache so that subsequent reads of the same data are served from memoryCorrect
   • CEvery read, including cache hits, must first pass through the oplog before being returned to the guest
   • DA cache miss always forces a remote read even when a local replica exists, because the cache only holds remote data
   ✓ Correct answer: A, B

   The unified cache is an in-memory read cache layered above the extent store. A cache hit returns data straight from RAM without any disk access, giving the lowest read latency. A miss reads from the local SSD extent store and then populates the cache with the data, so repeat reads of the same blocks become fast memory hits. This read-through caching behavior is what accelerates hot working sets. Both statements describe the standard Stargate read path correctly.

   Why the other options are wrong

   • CReads, including cache hits, do not pass through the oplog; oplog is a write buffer, and cache hits are served directly from memory.
   • DA cache miss reads from the local replica when one exists rather than forcing a remote read; the cache is not limited to holding only remote data.
5. Question 5Advanced Data Protection and Disaster Recovery

   An architect must satisfy an RTO of 30 minutes for a 50-VM application while keeping replication costs low. The team proposes hourly Async replication and a Recovery Plan with extensive per-VM NGT readiness scripts and long inter-stage delays. A tabletop test shows the orchestrated recovery alone consumes 55 minutes, breaching RTO, even though RPO is met. What design change BEST reduces RTO without weakening data protection?

   • AParallelize independent tiers into the same stage and trim unnecessary inter-stage delays/scripts so the Recovery Plan executes within the RTO, since RTO is dominated by orchestration time, not replication frequencyCorrect
   • BSwitch the application from hourly Async to NearSync replication, because tightening the RPO down to minutes automatically shortens the orchestrated recovery execution time and brings the total failover comfortably back under the RTO
   • CIncrease the Async replication frequency from hourly to every fifteen minutes so that the recovered VMs at the target are able to power on measurably faster during the orchestrated failover boot sequence
   • DAdd substantially more snapshot retention to the protection policy so that the Recovery Plan has many additional recovery points and can select a faster-booting one from among them during the failover
   ✓ Correct answer: A

   The tabletop shows RPO is met but the orchestration alone takes 55 minutes, so the RTO breach comes from how long the Recovery Plan runs, not how often data is replicated. Collapsing independent tiers into the same stage so they boot in parallel, and removing unnecessary inter-stage delays and readiness scripts, shortens execution to fit the 30-minute RTO. This reduces RTO without touching replication, so data protection is unchanged. The fix targets the actual bottleneck, which is orchestration time.

   Why the other options are wrong

   • BNearSync improves RPO (data currency), not orchestration duration; the 55-minute breach is caused by staging and delays, not RPO.
   • CMore frequent replication does not make VMs boot faster during failover; the RTO breach is in orchestration, not data freshness.
   • DExtra retention gives more recovery points but does not reduce the time the ordered stages and delays take to execute.
6. Question 6Security and Compliance

   After enabling Cluster Lockdown on a production cluster, an administrator reports that an automation script using stored username and password credentials over SSH to a CVM now fails with permission denied, although Prism web logins still work. What is the MOST likely root cause?

   • APrism Central revoked the cluster's registration token when lockdown was enabled
   • BCluster Lockdown disabled password-based SSH authentication, so only key-based SSH sessions are acceptedCorrect
   • CThe Cassandra ring lost quorum, blocking all credential validation
   • DSCMA reset the admin password to a random value during its scheduled run
   ✓ Correct answer: B

   Cluster Lockdown turns off password-based SSH to the CVMs while leaving Prism web logins and key-based SSH working. An automation script that authenticates over SSH with a stored username and password therefore fails with permission denied after lockdown, because password auth is no longer accepted. Prism web logins still work because lockdown targets SSH, not the web interface. Converting the script to use an authorized SSH key pair resolves the failure.

   Why the other options are wrong

   • APrism Central does not revoke the cluster's registration token when lockdown is enabled, and Prism logins still work, so registration is not the issue.
   • CA Cassandra quorum loss would affect cluster operations broadly, not selectively break password SSH while web logins succeed, so it is not the cause.
   • DSCMA hardens the OS baseline and does not randomly reset the admin password to break SSH; lockdown disabling password auth is the actual cause.
7. Question 7Advanced Monitoring and Troubleshooting

   A four-node cluster shows one node with controller latency 3x higher than its peers for the same VM I/O profile. CPU and memory on that CVM are normal, but its SSD wear/utilization differs. Which methodical next step BEST isolates whether the disk is the bottleneck?

   • AExamine per-disk latency and queue depth on that node's SSDs in the Hardware/Disk analysis charts to find a failing or saturated deviceCorrect
   • BImmediately reboot the CVM on that node to clear the elevated controller latency and force its services to restart cleanly
   • CDisable deduplication across the whole cluster to reduce the metadata processing load
   • DTrigger a manual Curator full scan so that data is rebalanced away from the slow node and its SSDs see less I/O pressure
   ✓ Correct answer: A

   One node showing three times the controller latency with normal CPU and memory but differing SSD wear points to a disk-level problem on that node. The methodical isolation step is to drill into per-disk latency and queue depth for that node's SSDs in the Hardware and Disk analysis charts, which reveals whether a specific device is failing or saturated. That per-device view confirms or rules out the disk as the bottleneck. It directly tests the hypothesis suggested by the differing SSD utilization.

   Why the other options are wrong

   • BRebooting the CVM is disruptive and skips diagnosis; it neither confirms nor isolates whether the SSD is the bottleneck.
   • CCluster-wide dedup changes are a heavy action unrelated to isolating a single node's disk latency.
   • DA Curator full scan moves data but does not identify whether a specific SSD is failing or saturated.
8. Question 8Lifecycle, Upgrades, and Migration

   A rolling AOS upgrade stalls after the first node. The LCM/upgrade log indicates it is waiting for the cluster to reach a state where it can tolerate one node failure, but the condition is never satisfied. Storage usage is moderate. Which underlying cluster condition is the MOST likely blocker?

   • AA disk or node in the cluster is already degraded or down, so resiliency cannot be confirmed before taking the next nodeCorrect
   • BPrism Central was unregistered from the cluster earlier, so the rolling upgrade cannot verify single-node fault tolerance state
   • CThe cluster is running too many idle guest VMs, which blocks the upgrade from confirming it can tolerate one node failure
   • DFoundation must be upgraded to a newer release before the AOS rolling upgrade can proceed past the first node
   ✓ Correct answer: A

   A rolling AOS upgrade only advances when it can confirm the cluster will still tolerate a node failure after the next node goes down. If a disk or node is already degraded or down, the cluster cannot reach that fully resilient state, so the upgrade waits indefinitely for a condition that never becomes true. Moderate storage usage rules out a capacity cause. Repairing the degraded component restores resiliency and lets the upgrade continue.

   Why the other options are wrong

   • BPrism Central registration is a management-plane concern; a rolling AOS upgrade evaluates data resiliency at the cluster level regardless of PC registration.
   • CIdle VMs consume no data-path resources and have no bearing on whether the cluster can rebuild after a node is taken offline.
   • DFoundation is an imaging tool that does not gate an in-progress AOS rolling upgrade waiting on resiliency confirmation.
9. Question 9Advanced AHV and VM Management

   An administrator must capture a crash-consistent point-in-time copy of a single running AHV VM for a one-time backup test, without installing any third-party backup software and without quiescing the guest. Which native mechanism is the most direct way to accomplish this from Prism?

   • ATake a VM-level snapshot from the VM's Manage menu in Prism ElementCorrect
   • BClone the VM and power on the clone to a new name
   • CExport the VM's vdisks to an OVA file using Prism Central image management
   • DConfigure a Protection Domain with a one-hour async schedule and wait for the first snapshot
   ✓ Correct answer: A

   AHV supports on-demand VM snapshots directly from Prism Element. Without application or guest agents to quiesce I/O, the snapshot is crash-consistent, which is exactly what the requirement specifies. It is a single, immediate action and requires no additional software, making it the most direct method.

   Why the other options are wrong

   • BCloning creates a full independent VM rather than a point-in-time copy, consumes more resources, and is not a backup/snapshot mechanism.
   • COVA export captures the VM image but is a slower, file-export operation intended for portability, not an immediate point-in-time backup test of a running VM.
   • DA Protection Domain with a schedule introduces ongoing replication policy and waiting for the schedule, which is far more than the one-time, immediate snapshot the requirement calls for.
10. Question 10Advanced Data Protection and Disaster RecoverySelect all that apply

    An architect is comparing Nutanix DR (Leap) entity-centric protection with legacy Protection Domains (PD) for a greenfield AHV deployment managed by Prism Central. Which TWO statements correctly describe advantages of the entity-centric Nutanix DR / Leap model over legacy Protection Domains?

    • AProtection is defined by categories so VMs are automatically protected based on tags rather than being manually added to a consistency groupCorrect
    • BOrchestrated, multi-stage failover with network mapping and boot ordering is provided through Recovery PlansCorrect
    • CReplication is configured per individual disk LUN rather than per VM
    • DProtection Domains support cross-AZ failover to AWS while Leap does not
    ✓ Correct answer: A, B

    Entity-centric protection in Prism Central uses categories so any VM matching a tag is automatically included in a Protection Policy, removing manual consistency-group maintenance. Recovery Plans then add orchestration: staged boot order, network mappings, IP remap, and in-guest scripts. Legacy PDs are entity-group based and lack this Prism Central orchestration.

    Why the other options are wrong

    • CNutanix DR protects at the VM level, not per LUN; this is not an advantage of Leap.
    • DIt is the opposite: Nutanix DR/Leap supports failover to NC2 on AWS, whereas legacy Protection Domains do not provide that cloud orchestration.

Related Nutanix resources

• NCM-MCI 6.10 study guideKey concepts
• Nutanix practice examsAll Nutanix
• Certification pathWhere this fits
• Certification exam guides & tipsBlog
• Plans & pricingFree & paid
• NCP-MCI 7.5 practice examRelated
• NCA 7.5 practice examRelated

NCM-MCI 6.10 practice exam FAQ