Azure VM Boot performance: batch 20 vs 50 VMs - should I start all at once?
I’m testing the startup time of around 100 Azure VMs using Azure CLI (az vm start). So far I have two measurements:
Batch size 20: ~30 minutes total for all 100 VMs
Batch size 50: ~46 minutes total for all 100 VMs
The important part is that these are total wall-clock times for all 100 VMs, not the time for a single batch
Increasing the concurrency from 20 to 50 actually made the overall operation significantly slower. I’m wondering if this could be caused by Azure ARM/Compute throttling, backend queuing, or some other limitation around VM power operations.
My next test will be to submit start requests for all 100 VMs concurrently, without waiting for each batch to complete, and then poll the VM power states until all machines are running.
I don’t have the result for the 100-at-once test yet.
Questions
- Has anyone tested starting 100+ Azure VMs simultaneously with Azure CLI?
- Is there a recommended concurrency/batch size?
- Can Azure VM power operations be throttled even if I’m not getting obvious 429 responses?
I’m specifically interested in real-world experience with large-scale Azure VM start operations.
I’ll update the post once I have the 100 at-once measurement.