The tool is a command-line utility that you run inside compatible Azure VMs. It is backed by Azure Boost. It provides remote disk I/O and latency metrics, and it maps each disk in the VM to its portal counterpart. Microsoft's Azure Compute blog post is dated November 17, 2026. It is not a new disk type, and it does not make storage faster. It makes storage easier to inspect.
What the tool is
The tool is called Azure-Boost-Observability. It supports both Windows and Linux guest operating systems, and users can modify it. The GitHub repository describes the Windows version as a utility for working with NVMe drives on Azure Boost-enabled VMs. It can query performance statistics and identify NVMe namespaces. A pull request in that repository also adds ARM64 build configurations. That matters because several of the supported VM series run on Cobalt Arm processors.
The post lists "1.0" as the version and November 16, 2026 as the last update. It does not give installation steps or command syntax. Microsoft points readers to a separate documentation page and the GitHub repository for those. I haven't verified either, so I won't invent a command line.
Metrics reported for remote disks
Microsoft lists these metrics for each remote disk:
- Operation counts: cumulative reads and writes completed over the disk's lifetime.
- Bytes transferred: cumulative bytes read and written over the disk's lifetime.
- Outstanding I/O: the number of requests currently outstanding on the disk.
- Latency: maximum and average latency for reads and writes.
- Dispatch latency: maximum and average dispatch latency for reads and writes.
- Hardware-acceleration state: a flag for when the system isn't running in hardware-accelerated mode.
Microsoft says outstanding I/O and latency use a 5-second sampling interval. It calls these the most granular remote-disk performance metrics on Azure. The post does not say the cumulative counters use that interval. Treat them as running totals, not per-5-second rates. If you want throughput over time, you would take samples and work out the difference yourself.
Matching guest disks to portal disks
Anyone who has chased a slow "disk 2" through several data disks knows the naming problem. For identification, the tool reports four things:
- Disk name.
- Role: operating system, data, local, or unknown.
- Size: the namespace size in GiB.
- NGUID: the Namespace Globally Unique Identifier.
The NGUID gives you a stable identifier to compare against what you see in the portal. Microsoft says this lets you associate performance metrics with the right disk.
Output formats
Per Microsoft, JSON output feeds scripts and monitoring, and text output suits quick investigation. The post does not name any specific monitoring integrations. It also doesn't mention alert thresholds, retention or an export destination. Treat those as things you would build around the tool yourself.
Compatibility and regional availability
You need a compatible Azure VM with NVMe disks attached. The currently compatible series are:
- Intel: Dlsv7, Dldsv7, Dsv7, Ddsv7, Esv7 and Edsv7.
- Cobalt 200: Dplsv7, Dpldsv7, Dpsv7, Dpdsv7, Epsv7, Epdsv7, Mpsv4, Mpdsv4 and Lpsv5.
Availability differs by feature:
| Feature | Scope today |
|---|---|
| Remote NVMe performance metrics | All Azure regions with compatible VM sizes |
| Disk identification (remote and local NVMe) | Currently in Early Updates Access Program (EUAP) regions |
Microsoft says availability in all regions and broader VM compatibility are planned for 2027. That is a plan, not a date. The post doesn't say which regions are in EUAP. If you test this in a production region, you may get performance metrics but no disk identification.
Azure's NVMe documentation adds more checks. NVMe support depends on VM size and OS image. Microsoft says it has no plans to support NVMe disks on Generation 1 VMs. A VM created from an image that isn't marked NVMe-capable ends up SCSI-based. So before blaming the tool, confirm three things: the series, the Generation 2 image, and that the disk controller type is NVMe.
Why it matters
Azure's built-in disk metrics have gaps for NVMe. Per the Azure disk metrics documentation, OS Disk Latency (Preview) is only available for SCSI-attached disks, not NVMe. A community Azure disk-observability project says the same about its Data Disk Latency metric. This tool partly fills that gap from inside the guest. As the NVMe-only v6 and v7 families spread, that gap matters more.
These are my practical readings of the metrics, not Microsoft claims:
- A rising outstanding-I/O count alongside rising latency suggests requests are queuing.
- A hardware-acceleration flag helps you rule out a VM running outside the accelerated path.
- Latency measured inside the guest, together with the NGUID mapping, helps you decide which managed disk to look at, or whether to look at the VM at all.
Caveats
- It diagnoses nothing automatically. It reports numbers and identity. Interpreting them is still your job.
- The scope is narrow. The performance metrics cover remote disks. Identification covers remote and local disks, but only in EUAP regions for now.
- Setup details are not in the announcement. Prerequisites beyond the VM series and NVMe attachment are not spelled out.
- It is a 1.0 release. Microsoft makes no performance claims for the tool itself, and none should be assumed.
Bottom line
For admins running supported v7 or Cobalt 200 VMs, this is a useful, low-risk addition. It gives fine-grained, in-guest remote-disk metrics and a way to map disks back to the portal. Check your VM series and region first, since disk identification is the part with the narrowest rollout.
References
- Announcing: Azure Boost observability for NVMe disks Azure Updates · 2026-11-18T00:00:09Z
- Added ARM64 configurations. by Ghinghes · Pull Request #2 · microsoft/Azure-Boost-Observability · GitHub github.com
- azure-compute-docs/articles/virtual-machines/enable-nvme-faqs.yml at main · MicrosoftDocs/azure-compute-docs · GitHub github.com