Azure VM Disk Performance Troubleshooting: IOPS and Throughput

Separate VM limits from disk limits

Azure VM sizes have limits for IOPS and throughput, while individual disks have their own performance characteristics. The effective workload limit can be constrained by either side of the path.

Identify the affected disk

Map the application volume to its Azure managed disk and operating-system device. Confirm whether the issue affects one disk or the entire VM.

Measure the workload

Look at read/write IOPS, throughput and latency together. A workload can hit an IOPS ceiling without reaching a throughput ceiling, or the reverse.

Check the VM size

If the VM size has lower storage bandwidth than the disks can provide, changing only the disk type may not solve the problem. Compare the workload requirement with the VM and disk limits.

Check application behavior

Large sequential transfers and small random I/O have different performance profiles. Review database settings, backup jobs, antivirus activity and scheduled scans that may change the workload during the incident.

Validate after changes

Capture the same performance counters before and after the change. This gives you evidence that the bottleneck moved or was removed.

Useful commands

Get-PhysicalDisk
Get-Counter '\LogicalDisk(*)\Avg. Disk sec/Read'
Get-Counter '\LogicalDisk(*)\Avg. Disk sec/Write'

Production checklist: capture the original state, test one dependency at a time, make the smallest controlled change and validate the original symptom before closing the incident.

What good troubleshooting looks like

Use evidence before configuration changes. Record the symptom, test result, change made and validation result so another engineer can repeat the procedure.

Workflow
Symptom → hypothesis → direct test → controlled change → validation → documentation

Frequently asked questions

What should I check first?

Start with the exact failed path or dependency and test it directly before changing configuration.

Should I change production configuration immediately?

No. Capture the current state first and make one controlled change at a time.

How should I document the fix?

Record the symptom, commands, result, configuration change and validation result.

Related TechRunbook guides

Primary reference

This guide was independently written for TechRunbook. Technical details were checked against current Microsoft documentation. Microsoft Learn reference →

Need more infrastructure runbooks?

Explore the TechRunbook article library for Windows Server, VMware, Hyper-V, Azure, PowerShell and MABS troubleshooting.

Browse all articles →