Troubleshooting Mindset Series #012 VM Slow but CPU & Memory Look Normal — Where Should We Look Next?

A VM suddenly becomes slow, but when checking the basic resource usage, CPU and memory are still within normal range.

So where should we look next?

This is where storage performance can easily be overlooked.

What I Check First

Before making any changes to the VM, I would first confirm whether the problem is isolated to the affected VM.

Check:

• CPU usage

  
  
• Memory usage

  
  
• Disk usage

  
  
• Disk I/O activity

  
  
• Disk latency

  
  
• Application response

  
  
  
  

If CPU and memory are normal but the application is still slow, don't immediately assume that the application is the problem.

Troubleshooting Steps

1. Check the affected VM

Look at the VM's disk activity and latency. A VM can have low CPU utilization while still experiencing high storage latency.

2. Compare with a healthy VM

If another VM on the same HCI environment is working normally, compare the two VMs.

Check their:

• VM configuration

  
  
• Disk configuration

  
  
• Storage location

  
  
• I/O activity

  
  
• Performance metrics

  
  
  
  

This comparison can quickly tell us whether the issue is VM-specific or infrastructure-related.

3. Check the HCI storage layer

If several VMs are experiencing similar symptoms, move the investigation toward the HCI storage layer.

Check for abnormal storage latency, high I/O workload, disk health, capacity, and any ongoing storage-related tasks.

Common Mistake

One common mistake is to immediately increase the VM's CPU or RAM just because users report that the application is slow.

If the actual bottleneck is storage I/O, adding more CPU or memory may not solve the problem.

Practical Takeaway

When troubleshooting VM performance, don't judge the problem from CPU and memory alone.

Start from the affected VM, compare it with a healthy VM, then move toward the HCI infrastructure if the evidence points there.

VM → Disk I/O → Storage → HCI

That approach helps narrow down the actual bottleneck before making unnecessary changes.

Thanks to share!