Explanation
Why the waste happens and who it affects.
On a regular flexible shape they are charged for every configured OCPU every hour, even though average use is a small fraction of it. OCI lets VM.Standard3.Flex, VM.Standard.E3.Flex, VM.Standard.E4.Flex and VM.Standard.E5.Flex instances run as burstable instances with a baseline of 12.5% or 50% of each OCPU, billed at the baseline, while still being able to burst to the full OCPU count.
Oracle's Cloud Advisor has a dedicated Change Compute Instances to Burstable recommendation for this pattern, and it calculates savings directly as the baseline fraction: moving to a 12.5% baseline costs 12.5% of the regular OCPU charge. The lever is different from rightsizing, because the instance keeps its OCPU count and memory, so the change suits workloads whose occasional peaks rule out a smaller shape.
Billing model
The pricing dimensions that drive this cost.
Burstable and regular instances on the same flexible shape differ only in how OCPUs are charged.
- Regular flexible instance
- Charged for the total configured OCPU count every hour, even if usage is lower
- Burstable baseline OCPU
- Charged at 12.5% or 50% of each configured OCPU, the same whether usage is below, at or above the baseline
- Memory
- Charged on the amount configured, the same as a regular instance; memory does not burst
- Windows Server license
- Charged according to the baseline OCPU on burstable instances
How to detect
4 checks to find it in your estate.
- Review Cloud Advisor's Change Compute Instances to Burstable recommendation: it checks VM.Standard3.Flex, E3.Flex and E4.Flex instances over 24 hours (48 CPU data points) and flags those with 5 or fewer points above 50% or 12.5% Max CPU, P95 CPU below 40% or 10%, and a memory-to-core ratio under 12 GB per OCPU
- For shapes and patterns Cloud Advisor does not cover, such as VM.Standard.E5.Flex, chart CpuUtilization (Max and P95) in the oci_computeagent namespace over several weeks to find instances with low sustained CPU and short spikes
- Prioritize non-production compartments and workload types Oracle lists as burstable fits: development and test, CI/CD, monitoring, microservices and static websites
- Exclude instances that use features burstable instances do not support: dedicated virtual machine hosts, preemptible capacity, capacity reservations and extended memory VMs
How to fix
5 ways to remove the waste.
- Change eligible instances to a burstable configuration with Change shape, selecting a 12.5% or 50% baseline; Oracle recommends the baseline that matches the threshold the recommendation flagged
- Use the Cloud Advisor fix-it flow where available; it also switches the instance to paravirtualized networking, which burstable instances require, so plan for that change on instances using SR-IOV
- Schedule the change: a running instance is rebooted during a shape change, and secondary VNICs may need to be reconfigured afterwards
- Keep regular instances for production or CPU-critical workloads: bursts are limited to one continuous hour, the host is oversubscribed so bursting and maximum network bandwidth are not guaranteed, and memory does not burst
- After the change, watch CpuUtilization for sustained use at the baseline, which means the instance needs the 50% baseline or a regular configuration
Documentation
Vendor references for pricing and configuration.