Our First Azure Observations: Strong Compute, Emerging Network Signals
When Azure joined the Webbynode dataset last week, we weren't looking for winners or losers.
The goal was simply to begin building repeated observations from fresh deployments, just as we've done across AWS, Google Cloud, Linode, DigitalOcean, Vultr — and now Oracle Cloud Infrastructure (OCI).
So far we've completed our first observation windows across several Azure VM families and regions. It's still early — most clusters currently consist of only three fresh deployments — but a few interesting patterns are already beginning to emerge.
Explore the Live Azure Dataset
This article summarizes our initial observations. The underlying Azure benchmark dataset continues to grow and is updated as additional observation windows are completed.
→ View the live Azure Provider page
Strong Compute Consistency
One thing became obvious almost immediately: CPU performance has been remarkably repeatable.
Our latest Japan East cluster (Standard D16ds_v6) produced three deployments with extremely consistent sustained CPU throughput and storage performance. That's a premium-class VM, so a high degree of consistency isn't especially surprising. Modern cloud platforms have become very good at delivering predictable compute performance.
Increasingly, CPU is no longer the most interesting part of the benchmark.
The Network Tells a Different Story
Where things became more interesting was the network. Several Azure clusters showed measurable outbound throughput variation across our fixed measurement endpoints.
For example, our first Central US D8ds_v7 observation window showed noticeably different throughput behavior across repeated fresh deployments despite highly consistent CPU and storage results.
It's important to keep these observations in perspective. None of these clusters have enough observation depth to support broad conclusions. But they reinforce something we've now seen across multiple providers. Ultimately, infrastructure behavior isn't defined solely by CPU performance.
Routing, regional network paths, and deployment-to-deployment variability often become the more interesting observations.
Three Deployments Are Only The Beginning
One principle of Webbynode hasn't changed. Three fresh deployments establish an initial observation window. They do not establish long-term infrastructure behavior.
Some of the most interesting findings in the dataset have only appeared after repeating the exact same benchmark days — or even weeks — later. We've previously observed identical provider, region, and plan combinations producing dramatically different outcomes following infrastructure upgrades, routing changes, or transient network events.
That's why every cluster remains subject to future observation windows.
The value isn't just in running another benchmark, but in understanding how infrastructure behaves over time.
Azure Is Just Getting Started
The Azure dataset remains small compared to providers we've been observing for months.
Current coverage includes multiple VM families across several regions, with additional repeated deployments already planned. As those observation windows expand, we'll begin separating temporary events from repeatable infrastructure characteristics.
That's ultimately the objective.
Not to produce a faster benchmark.
Not to declare winners after three deployments.
But to build longitudinal evidence about how cloud infrastructure actually behaves over time.
That's infrastructure behavior intelligence.