An administrator is troubleshooting a storage performance issue on an ESXi 9.0 host where multiple virtual machines are experiencing high I/O latency. Which THREE actions can the administrator take to analyze and mitigate the latency? (Choose three.)
Trap 1: Disable all host CPU power management states to accelerate storage…
CPU power management states govern processor idle latency, not storage driver interrupt handling; disabling them cannot reduce I/O latency caused by queue depth or path saturation. It is tempting because C-states and P-states do affect wake-up latency, but that mechanism applies to CPU-bound workloads, not storage I/O.
Trap 2: Format the ESXi boot bank with VMFS-6 to clear storage queue…
VMFS-6 is a datastore filesystem, not a boot bank format; the ESXi boot bank uses FAT16, and reformatting it would destroy the hypervisor rather than clear any storage queue. It is tempting because VMFS-6 supports advanced queueing features, but those apply to shared datastores, not the boot device.
- A
Disable all host CPU power management states to accelerate storage driver interrupt handling.
Why it fails: CPU power management states govern processor idle latency, not storage driver interrupt handling; disabling them cannot reduce I/O latency caused by queue depth or path saturation. It is tempting because C-states and P-states do affect wake-up latency, but that mechanism applies to CPU-bound workloads, not storage I/O.
- B
Format the ESXi boot bank with VMFS-6 to clear storage queue buffers.
Why it fails: VMFS-6 is a datastore filesystem, not a boot bank format; the ESXi boot bank uses FAT16, and reformatting it would destroy the hypervisor rather than clear any storage queue. It is tempting because VMFS-6 supports advanced queueing features, but those apply to shared datastores, not the boot device.
- C
Use esxtop ('d' view) to separate Kernel Latency (KAVG) from Device Latency (DAVG) to determine whether the bottleneck is internal to ESXi or on the storage array.
Using esxtop's 'd' view isolates the latency source: KAVG measures time spent in the VMkernel queue, while DAVG reflects the storage device's response. This distinction satisfies the stem's need to determine whether the bottleneck lies within ESXi or on the array, directing mitigation correctly.
- D
Verify multipathing policies (e.g., Round Robin) and ensure all physical paths to the storage array are active and healthy.
Multipathing determines how I/O is distributed across available paths to the array. Verifying Round Robin policy and confirming every physical path is active and healthy prevents saturation or failover-induced latency, directly mitigating the reported bottleneck.
- E
Enable or configure Storage I/O Control (SIOC) to prevent single noisy virtual machines from saturating datastore queue depths.
Storage I/O Control enforces per-datastore queue-depth throttling, capping the I/O a single noisy virtual machine can issue so it cannot saturate shared datastore queues. This directly addresses the high I/O latency caused by contention across multiple virtual machines on the ESXi 9.0 host.