Disk Performance Issues¶

1. Summary¶

Symptom¶

Disk latency rises, queue depth builds, or workloads hit IOPS and throughput limits even though the VM itself is otherwise reachable.

Why this scenario is confusing¶

The limit may exist on the individual disk, the VM SKU aggregate cap, or the caching and workload pattern rather than the disk tier alone.

Troubleshooting decision flow¶

graph TD
    A[Disk performance issue] --> B{VM cap exceeded?}
    B -->|Yes| C[Upgrade VM size]
    B -->|No| D{Disk cap or burst exhausted?}
    D -->|Yes| E[Resize or change tier]
    D -->|No| F[Check caching and workload profile]

2. Common Misreadings¶

"Premium SSD always removes latency problems."
"Only disk tier matters."
"Caching should be enabled everywhere."

3. Competing Hypotheses¶

H1: VM-level aggregate IOPS or throughput cap reached.
H2: Disk-level cap or burst exhausted.
H3: Wrong host caching mode for workload.
H4: Guest workload pattern is creating queue saturation.

4. What to Check First¶

Disk type, size, and expected performance ceiling.
VM SKU storage limits.
Read/write latency, IOPS, throughput, and queue depth.
Host caching configuration and workload role.

5. Evidence to Collect¶

Azure Monitor disk metrics and iostat or Perfmon output.
Whether issue is read-heavy, write-heavy, or mixed.
Current caching settings on OS and data disks.
Whether Ultra Disk or Premium SSD v2 is involved.

6. Validation and Disproof by Hypothesis¶

H1: VM-level cap reached¶

Supports: sum of disk demand approaches VM limit.
Weakens: plenty of VM headroom remains.

H2: Disk-level cap reached¶

Supports: one disk hits its known IOPS/throughput boundary.
Weakens: the disk is below its service limit.

H3: Wrong caching mode¶

Supports: logs/data files use unsuitable cache mode and latency varies.
Weakens: Ultra Disk or Premium SSD v2 where host caching does not apply.

H4: Guest workload queue saturation¶

Supports: guest queue grows before Azure-side limits are hit.
Weakens: guest queue normal while Azure throttling is explicit.

7. Likely Root Cause Patterns¶

Premium disk attached to an undersized VM.
Standard SSD burst depleted during heavy activity.
Database log disk left on cached mode.
Too many hot workloads share the same VM storage budget.

8. Immediate Mitigations¶

Move to a VM size with higher storage ceilings.
Resize or retier the hot disk.
Correct caching mode for data versus log paths.
Spread heavy I/O across disks where appropriate.

9. Prevention¶

Capacity-plan against both disk and VM limits.
Review caching per workload role.
Monitor queue depth and latency, not just throughput.