After a vertical scaling task is executed, you must restart the instances to put the new instance types into effect. During the restart period, the service availability may be compromised.
Vertical scaling lets you resize pay-as-you-go Elastic Compute Service (ECS) instances within a scaling group — upgrading to handle peak demand and downgrading during off-peak hours to cut costs. Auto Scaling automates these changes on a schedule or based on CloudMonitor metrics, eliminating manual intervention.
Use cases
Vertical scaling suits two common workload patterns:
Predictable peaks: Your workload spikes follow a known schedule — tasks at the beginning of each month, payroll processing, end-of-quarter reporting, or promotional events. Create a scheduled vertical scaling task to upgrade instances before the peak and downgrade when it ends.
Variable workloads: Traffic fluctuates unpredictably — a news platform that cannot anticipate breaking stories, a B2B service with uneven API usage, or a SaaS application with irregular user activity. Create a CloudMonitor metric-based vertical scaling task. Auto Scaling monitors the metric you specify and resizes instances automatically when values cross your threshold.
How vertical scaling works
Auto Scaling handles the entire resize sequence without manual steps:
Stops the instance.
Changes the instance type.
Restarts the instance.
The instance billing rate adjusts after the type change: rates increase after an upgrade and decrease after a downgrade.
Benefits
Efficiency — Automatically upgrades or downgrades ECS instance types in a scaling group, saving you from creating or releasing instances and making custom configurations.
Convenience — No manual steps needed. Auto Scaling handles stopping the instance, changing the instance type, and restarting the instance on your behalf.
Flexibility — Supports both scheduled adjustment and CloudMonitor metric-based adjustment of instance types, so you can choose the approach that fits your workload.
Cost-effectiveness — Downgrade instance types during off-peak hours to reduce resource costs, and reduce O&M costs by eliminating the need to continuously monitor resource usage.
Limitations
| Constraint | Details |
|---|---|
| Instance billing type | Pay-as-you-go instances only |
| Instance family | Must stay within the same instance family (e.g., ecs.g6.large to ecs.g6.2xlarge, not to ecs.c6.2xlarge or ecs.r6.2xlarge) |
| Task direction | Each task handles either upgrade or downgrade, not both — create separate tasks for each direction |
| Upgrade task: max instance types | Up to 10 instance types in ascending order of computing power (vCPUs and memory) |
| Downgrade task: max instance types | Up to 5 instance types in descending order of computing power (vCPUs and memory) |
| Recurring task interval | Greater than 30 minutes between consecutive cycles — shorter intervals cause frequent restarts that affect service availability |
Billing
Vertical scaling itself is free. Instance charges adjust after each type change — upgrade increases the rate, downgrade decreases it. For pay-as-you-go pricing details, see Pay-as-you-go.