Explanation
Why the waste happens and who it affects.
Base ECPUs are often sized for a peak, a load test or a migration and left there, so Autonomous AI Lakehouse (formerly ADW) and Autonomous AI Transaction Processing (ATP) databases run for months with average CPU far below what is allocated. Oracle's Cloud Advisor flags this case with a dedicated cost management recommendation.
Compute auto scaling makes a high base unnecessary for many workloads. With auto scaling enabled, a database can use up to three times its base ECPU count when the workload needs it and is billed for the extra ECPUs only for the time they are used. Compute auto scaling is enabled by default on new databases, so a lower base can still absorb peaks while the base charge drops for every hour.
Billing model
The pricing dimensions that drive this cost.
ECPU usage is measured each second in whole ECPUs and averaged across each hour, with a one-minute minimum; storage is billed separately.
- Base ECPUs
- The ECPU count selected for the database, billed every hour the database is running whether or not it is used
- Auto-scaled ECPUs
- Additional ECPUs used above the base when compute auto scaling is enabled, up to 3x the base, billed only for the time consumed
- Stopped database
- A stopped Autonomous AI Database has zero ECPU usage, while storage and backups continue to bill
- Local standby ECPUs
- A local Autonomous Data Guard peer adds the primary's base ECPUs again, so an oversized base is paid twice
How to detect
5 checks to find it in your estate.
- Review Cloud Advisor's Downsize Underutilized ADW and ATP Databases recommendation, which flags Lakehouse and Transaction Processing databases whose average CPU utilization over the last 30 days is below 30% of allocated CPUs
- Chart CpuUtilization and ECPUsAllocated in the oci_autonomous_database metric namespace for each database over a representative period, including month-end or batch peaks
- Use the Number of ECPUs allocated graph on the Database Dashboard in Database Actions to see average hourly ECPU use against the base count
- List databases with compute auto scaling disabled and a high base ECPU count, since they pay for peak capacity every hour
- Check databases with a local Autonomous Data Guard standby, where the base ECPU count is billed on both primary and peer
How to fix
5 ways to remove the waste.
- Lower the base ECPU count from the console (Manage resource allocation) or API, using the Cloud Advisor suggestion as a starting point; the Cloud Advisor fix-it flow reduces cores by 20 percent at a time
- Enable compute auto scaling on databases with bursty load so peaks are served by auto-scaled ECPUs billed per second instead of a permanently high base
- Stop Autonomous databases that are not needed for periods of time, since stopped databases have zero ECPU usage
- Tradeoff: the tp and tpurgent service concurrency limits scale with the ECPU count (75 x number of ECPUs), and the high and medium limits depend on whether auto scaling is enabled, so check peak concurrent sessions before lowering the base
- Re-check CpuUtilization after each change and repeat in steps rather than cutting the base in one move
Documentation
Vendor references for pricing and configuration.
