Azure Cost Management Tips: Deallocate, License, Commit

Azure cost management is the practice of tracking, allocating, and reducing Azure spend, using Microsoft Cost Management for visibility and a set of Azure-specific levers for reduction. The tips below are ordered from the ones that remove waste to the ones that discount what is left, and each one states the billing rule it depends on.

1. Deallocate, do not just stop

Azure distinguishes two “off” states for a VM. Stopped (allocated), the result of the PowerOff API or a shutdown from inside the guest OS, keeps the VM on its host and is still billed for compute; the REST reference for power off says you are still charged for the VM. Stopped (deallocated) releases the hardware lease, and compute is not billed. In both states some resources, such as disks and networking, continue to incur charges.

So a shutdown script inside Windows or Linux does not save compute cost on its own. Use the Deallocate operation, or a tool that calls it. The built-in VM auto-shutdown feature can shut VMs down daily and send a notification first; note that it defaults to UTC unless you set the time zone.

2. Use the serverless tier where a database is idle

Azure SQL Database has no general stop. Its serverless compute tier bills compute per second and can auto-pause a database after a configurable inactivity delay, after which only storage is billed; it resumes on the next login or activity. Auto-pause is supported only in the General Purpose service tier. For provisioned databases, the lever is sizing, not stopping.

3. Apply licences you already own

Azure Hybrid Benefit lets you use on-premises Windows Server core licences with active Software Assurance, or qualifying subscription licences, to run Windows VMs at a reduced cost, for new VMs or by updating existing ones. It is a licensing setting on the VM, not a change to its size or state.

4. Choose the right commitment

• Reservations commit to one or three years for a specific product and can reduce cost by up to 72% from pay-as-you-go prices. They are a billing discount that applies automatically to matching resources and do not change their runtime state. Reservations can be exchanged, and refunded within a limit of $50,000 USD in a rolling 12-month window.
• Savings plan for compute commits to an hourly spend. Each hour the benefit applies to eligible usage, starting with the usage that gets the highest discount; usage above the commitment bills at pay-as-you-go rates, and unused commitment for an hour expires.

Because unused hourly commitment expires, size either commitment to the capacity that runs around the clock after you have scheduled the rest. See reserved instances and savings plans.

5. Know what the AKS tier costs

AKS offers Free, Standard, and Premium tiers for cluster management. The Free tier has free cluster management and no financially backed uptime SLA, and is recommended for development and test clusters with fewer than 10 nodes. The Standard tier adds a financially backed uptime SLA at $0.10 per cluster per hour (Azure Retail Prices, East US), and Premium adds Long Term Support on top. Scaling node pools down removes node cost; the tier fee is a separate line.

6. Other levers worth checking

• Azure Advisor recommends shutting down or resizing underutilized VMs and scale sets, based on CPU and outbound network utilization.
• Spot VMs can be evicted with 30 seconds’ notice when Azure needs the capacity; the default eviction policy deallocates the VM, which keeps disk charges, while Delete removes the disks too.
• Dev/Test pricing offers discounted rates for dev and test workloads to active Visual Studio subscribers.

Where ZopNight fits

ZopNight’s scheduler handles Azure’s provider-specific power operations, including deallocate versus stop, for Azure VMs, AKS, MySQL and PostgreSQL Flexible Server, and other supported types. A recommendation flags Windows VMs that could use Azure Hybrid Benefit, a reversible licensing change. Billing sync uses amortized cost for Azure so reservation and savings plan purchases are spread across the resources that consumed them. See scheduling Azure VMs.