Explanation
Why the waste happens and who it affects.
Instances are left running after a proof of concept ends, a test cycle finishes, a workload is migrated elsewhere or an owner leaves, and show almost no CPU, memory or network activity for weeks. Oracle's Cloud Advisor treats these separately from lightly loaded instances: its Delete Idle Compute Instances recommendation targets instances that are effectively unused and should be removed, not resized.
Two OCI-specific details make idle instances easier to miss. Shutting an instance down from inside the operating system does not stop billing; only a stop from the Console or API does, and only for shapes that pause billing when stopped. And instances on Dense I/O, most GPU, HPC and bare metal Optimized shapes keep billing even when stopped, because their local NVMe or GPU resources are preserved.
Billing model
The pricing dimensions that drive this cost.
Compute resources are billed at per-second granularity with a one-minute minimum, independent of utilization.
- OCPUs and memory
- Billed per OCPU-hour and per GB-hour for every second a running instance exists
- Stopped Standard instance
- Stopping from the Console or API pauses billing, but the instance still counts toward service limits
- Stopped Dense I/O, GPU or HPC instance
- Continues to bill while stopped because local NVMe or GPU resources are preserved (VM.GPU.A10 and VM Optimized shapes excepted)
- Boot volume
- Billed separately as block storage and retained by default when the instance is deleted
How to detect
5 checks to find it in your estate.
- Review Cloud Advisor's Delete Idle Compute Instances recommendation, which analyzes the past seven days: for supported VM shapes, CPU utilization P95 below 3%, memory utilization maximum below 8% and network utilization maximum below 3% (for bare metal, CPU P95 below 2%, memory maximum below 5%, network maximum below 3%)
- For shapes Cloud Advisor does not scan, chart CpuUtilization, MemoryUtilization, NetworksBytesIn and NetworksBytesOut in the oci_computeagent namespace over at least seven days
- Act on Cloud Advisor's Enable Monitoring on Compute Instances recommendation first: without the Compute Instance Monitoring plugin, utilization data is not available to Cloud Advisor
- Look for instances that are running according to OCI but were shut down inside the guest OS, and for stopped instances on Dense I/O, GPU or HPC shapes that are still billing
- Check owner and environment tags and the compartment of each idle instance to find the team that can confirm it is unused
How to fix
4 ways to remove the waste.
- Confirm with the owner that the instance is unused, then terminate it with the Cloud Advisor fix-it flow or from the Console or API
- Decide on the boot volume at termination: Cloud Advisor's deletion keeps the boot volume, so either take a boot volume backup or custom image and then delete the volume, or keep it deliberately; retained boot volumes keep billing as block storage
- If an instance on a Standard shape must be kept but is not used for now, stop it from the Console or API rather than from the OS to pause compute billing; for Dense I/O, GPU and HPC shapes, termination is the only way to stop billing
- Add owner and expiry tags to new instances and review idle instances on a schedule so that sandboxes and test environments are removed when their project ends
Documentation
Vendor references for pricing and configuration.