Elastically reserved compute units (CUs) let you dynamically reserve compute resources as needed. Evaluate your business needs and purchase elastically reserved CUs. Then, set the number of elastically reserved CUs in a quota plan to expand capacity during peak periods and reduce it when demand drops.
Prerequisites
Before you configure elastically reserved CUs, ensure that you have:
Subscription reserved compute resources — purchase subscription reserved compute resources first, as elastically reserved CUs extend an existing subscription quota and cannot be configured independently
An activated MaxCompute subscription instance — see Activate MaxCompute
How it works
Elastically reserved CUs belong to a Dedicated resource pool. They are optional and sit on top of your subscription CUs, expanding quota capacity as needed.
The basic unit is a CU: 1 CU = 1 CPU core + 4 GB memory.
Two ways to scale
Immediate scaling — Set the number of elastically reserved CUs in a single quota plan and apply it immediately. Use this approach when you need to expand capacity for a known workload.
Time-based scaling — Create multiple quota plans, each with a different CU count, and attach a time-based schedule to switch between them automatically. Use this approach when your workload follows a predictable pattern, such as batch jobs that run overnight or reporting queries that peak during business hours.
Both approaches are configured on the Quotas page of the MaxCompute console. For configuration steps, see Manage compute resources using quotas.
Limitations
The maximum number of elastically reserved CUs for a quota cannot exceed the number of subscription CUs purchased for that quota, up to a maximum of 10,000 CUs.
Each scale-out requires a minimum of 50 CUs and must increase in increments of 50.
Scale-outs are subject to inventory availability. The system checks inventory when you configure a scale-out.
Billing
Elastically reserved CUs are billed at the pay-as-you-go rate for compute fees. For pricing details, see Compute fees (Pay-as-you-go).
What's next
Manage compute resources using quotas — configure quota plans and time-based schedules
Use of computing resources — assign jobs to a quota and start running workloads