All Products
Search

This Product

    Elastic Compute Service:Reserved instance usage rules

    Document Center

    Elastic Compute Service:Reserved instance usage rules

    Last Updated:Sep 24, 2026

    Reserved Instance are a type of discount benefit you purchase. After purchase, they do not directly provide you with corresponding resources; instead, they must be used together with pay-as-you-go resources you have purchased. If you do not currently have any eligible pay-as-you-go instances to offset after purchase, the Reserved Instance Coupon will continue to be billed. Understanding the deduction rules for Reserved Instance Coupons can help you make better use of discount benefits and reduce operating costs. This chapter introduces the rules and examples for how Reserved Instance Coupons are matched with pay-as-you-go instances.

    Offset rules

    • You do not need to manually use reserved instances to offset pay-as-you-go instances.

    • After you purchase a reserved instance, it automatically matches eligible pay-as-you-go instances during its validity period. After a match is made, the reserved instance offsets the charges for instance compute resources every hour based on its computing power (the normalization factor of the instance type * the number of reserved instances) until the reserved instance expires.

      Note

      A normalization factor represents the relative computing power of an instance type and usually depends on the number of vCPUs. When you use a reserved instance, the instance type and its normalization factor determine how much computing power the reserved instance delivers. To view the normalization factor of each instance type, see View normalization factor.

    • Matching conditions for reserved instances and pay-as-you-go instances

      Regional reserved instances and zonal reserved instances have different matching conditions. The following attributes determine whether a reserved instance matches a pay-as-you-go instance.

      Attribute

      Regional reserved instance

      Zonal reserved instance

      Region and zone

      Can match pay-as-you-go instances across zones in the specified region.

      Can match only pay-as-you-go instances in the specified zone.

      Instance type

      • Within the specified instance family, a smaller reserved instance can match a larger pay-as-you-go instance, and a larger reserved instance can match a smaller pay-as-you-go instance.

      • Resource reservation is not supported.

      • Can match only pay-as-you-go instances of the same instance size.

      • Resource reservation is supported. During the validity period, the specified number of instances of the specified instance type is reserved to ensure that pay-as-you-go instances can always be created in the specified zone.

      Operating system platform

      Can match only pay-as-you-go instances with the same operating system platform.

      Based on the matching conditions, regional and zonal reserved instances can offset pay-as-you-go instances only when all of the following conditions are met:

      Regional reserved instances

      • The regions are the same.

        Regional reserved instances support offsetting pay-as-you-go instances across zones.

      • The instance families are the same.

        Regional reserved instances support the following: a smaller reserved instance offsets a larger pay-as-you-go instance, and a larger reserved instance offsets a smaller pay-as-you-go instance.

      Note

      When instance sizes differ, you can use computing power to evaluate reserved instance usage.

      • The operating system platforms are the same.

      Zonal reserved instances

      • The region and zone are the same.

      • The instance family and instance size are the same.

      • The operating system platforms are the same.

      Note

      After you purchase a reserved instance, you can also use the offset forecast feature to view the pay-as-you-go instances that the reserved instance can match. For more information, see Purchase a reserved instance.

    Items eligible for offset

    • Only the charges for compute resources (vCPUs and memory) of pay-as-you-go instances can be offset. Charges for network, storage, and other resources cannot be offset. For information about ECS instance billing, see Billing overview.

    • Windows reserved instances can also offset image charges.

      Note

      The purchase price of a reserved instance for Windows includes the cost of the Windows image. This cost can be used to offset the image portion of the bill for a pay-as-you-go instance that runs Windows.

    Offset priority

    Reserved instances and savings plans can both offset the bills for compute resources of pay-as-you-go instances. If you have multiple discount benefits, the offset priority, from highest to lowest, is as follows:

    1. Reserved instances and resource plans, such as storage capacity units (SCUs).

      Note

      A reserved instance (RI) offsets the bills for pay-as-you-go compute resources, and a storage capacity unit (SCU) offsets the bills for pay-as-you-go storage resources. Using RIs and SCUs together in pay-as-you-go scenarios reduces compute and storage costs.

    2. ECS compute savings plans.

    3. General-purpose savings plans.

    4. Coupons.

    Offset examples

    Regional reserved instance offset examples

    Examples of successful offsets

    Example scenario

    Regional reserved instance

    Pay-as-you-go instance

    Offset result

    One reserved instance offsets one instance (a smaller reserved instance offsets a larger instance)

    You have one active regional reserved instance with the following attributes:

    • Region and zone: China (Qingdao), across zones

    • Instance type: ecs.g5.xlarge

    • Operating system platform: Linux

    • Number of instances: 1

    You have one pay-as-you-go instance with the following configuration:

    • Region and zone: China (Qingdao), Zone B

    • Instance type: ecs.g5.2xlarge

    • Image: Linux

    1. View the normalization factor for each instance type.

      • The normalization factor of the reserved instance ecs.g5.xlarge is 4.

      • The normalization factor of the pay-as-you-go instance ecs.g5.2xlarge is 8.

    2. Compare computing power.

      • Reserved instance: one reserved instance delivers computing power of 4 per hour (1 instance * 4).

      • Pay-as-you-go instance: one instance consumes computing power of 8 per hour (1 instance * 8).

    3. Offset result: Based on the computing power comparison, one reserved instance can offset 50% of the bill for one pay-as-you-go instance per hour.

    Multiple reserved instances offset one instance (smaller reserved instances offset a larger instance)

    You have two active regional reserved instances:

    • One reserved instance has the following attributes:

      • Region and zone: China (Qingdao), across zones

      • Instance type: ecs.g5.xlarge

      • Operating system platform: Linux

      • Number of instances: 1

    • Another reserved instance has the following attributes:

      • Region and zone: China (Qingdao), across zones

      • Instance type: ecs.g5.xlarge

      • Operating system platform: Linux

      • Number of instances: 1

    You have one pay-as-you-go instance with the following configuration:

    • Region and zone: China (Qingdao), Zone B

    • Instance type: ecs.g5.2xlarge

    • Image: Linux

    1. View the normalization factor for each instance type.

      • The normalization factor of the reserved instance ecs.g5.xlarge is 4.

      • The normalization factor of the pay-as-you-go instance ecs.g5.2xlarge is 8.

    2. Compare computing power.

      • Reserved instances: two reserved instances deliver computing power of 8 per hour (2 instances * 4).

      • Pay-as-you-go instance: one instance consumes computing power of 8 per hour.

    3. Offset result: Based on the computing power comparison, two reserved instances can offset only 100% of the bill for one pay-as-you-go instance per hour.

    One reserved instance offsets one instance (a larger reserved instance offsets a smaller instance)

    You have one active regional reserved instance with the following attributes:

    • Region and zone: China (Qingdao), across zones

    • Instance type: ecs.g5.4xlarge

    • Operating system platform: Linux

    • Number of instances: 1

    You have one pay-as-you-go instance with the following configuration:

    • Region and zone: China (Qingdao), Zone B

    • Instance type: ecs.g5.2xlarge

    • Image: Linux

    1. View the normalization factor for each instance type.

      • The normalization factor of the reserved instance ecs.g5.4xlarge is 16.

      • The normalization factor of the pay-as-you-go instance ecs.g5.2xlarge is 8.

    2. Compare computing power.

      • Reserved instance: one reserved instance delivers computing power of 16 per hour (1 instance * 16).

      • Pay-as-you-go instance: one instance consumes computing power of 8 per hour (1 instance * 8).

    1. Offset result: Based on the computing power comparison, one reserved instance can offset 100% of the bill for one pay-as-you-go instance per hour, and 50% of the reserved instance remains unused.

    One reserved instance offsets multiple instances (a larger reserved instance offsets smaller instances)

    You have one active regional reserved instance with the following attributes:

    • Region and zone: China (Qingdao), across zones

    • Instance type: ecs.g5.4xlarge

    • Operating system platform: Linux

    • Number of instances: 1

    You have four pay-as-you-go instances.

    • Two instances have the following configuration:

      • Region and zone: China (Qingdao), Zone B

      • Instance type: ecs.g5.xlarge

      • Image: Linux

    • Two instances have the following configuration:

      • Region and zone: China (Qingdao), Zone C

      • Instance type: ecs.g5.xlarge

      • Image: Linux

    1. View the normalization factor for each instance type.

      • The normalization factor of the reserved instance ecs.g5.4xlarge is 16.

      • The normalization factor of the pay-as-you-go instance ecs.g5.xlarge is 4.

    2. Compare computing power.

      • Reserved instance: one reserved instance delivers computing power of 16 per hour (1 instance * 16).

      • Pay-as-you-go instances: four instances consume computing power of 16 per hour (4 instances * 4).

    3. Offset result:

      • The pay-as-you-go instances are in two zones, but the regional reserved instance supports zone flexibility, which does not affect the offset.

      • Based on the computing power comparison, one reserved instance offsets 100% of the bills for four pay-as-you-go instances per hour.

    Examples of unsuccessful offsets

    Example scenario

    Regional reserved instance

    Pay-as-you-go instance

    Offset result

    Offset fails

    You have one active regional reserved instance. The reserved instance has the following attributes:

    • Region and zone: China (Qingdao), across zones

    • Instance type: ecs.g5.4xlarge

    • Operating system platform: Linux

    • Number of instances: 1

    You have one pay-as-you-go instance. The instance has the following configuration:

    • Region and zone: China (Qingdao), Zone B

    • Instance type: ecs.g5.xlarge

    • Image: Windows

    The reserved instance fails to match the pay-as-you-go instance because their operating system platforms differ. Therefore, the reserved instance cannot offset the bill for the pay-as-you-go instance.

    Offset fails

    You have one active regional reserved instance. The reserved instance has the following attributes:

    • Region and zone: China (Qingdao), across zones

    • Instance type: ecs.g5.xlarge

    • Operating system platform: Linux

    • Number of instances: 1

    You have one pay-as-you-go instance. The instance has the following configuration:

    • Region and zone: China (Hangzhou), Zone B

    • Instance type: ecs.c5.xlarge

    • Image: Linux

    The reserved instance fails to match the pay-as-you-go instance for the following reasons:

    • The reserved instance and the pay-as-you-go instance are not in the same region.

    • The instance types of the reserved instance and the pay-as-you-go instance do not belong to the same instance family, so the reserved instance cannot offset the bill for the pay-as-you-go instance.

    Zonal reserved instance examples

    Examples of successful offsets

    Example scenario

    Zonal reserved instance

    Pay-as-you-go instance

    Offset result

    Successful offset (one reserved instance offsets one instance)

    You have one active zonal reserved instance with the following attributes:

    • Region and zone: China (Qingdao), Zone B

    • Instance type: ecs.g5.xlarge

    • Operating system platform: Windows

    • Number of instances: 1

    You have one pay-as-you-go instance with the following configuration:

    • Region and zone: China (Qingdao), Zone B

    • Instance type: ecs.g5.xlarge

    • Image: Windows

    1. View the normalization factor for each instance type. The reserved instance and the pay-as-you-go instance have the same instance type and therefore the same normalization factor. The computing power of ecs.g5.xlarge is 4.

    2. Compare computing power.

      • Reserved instance: one reserved instance delivers computing power of 4 per hour (1 instance * 4).

      • Pay-as-you-go instance: one instance consumes computing power of 4 per hour.

    3. Offset result: The computing power is the same. Therefore, one reserved instance can offset 100% of the bill for one pay-as-you-go instance per hour.

    One reserved instance offsets multiple instances

    You have one active zonal reserved instance with the following attributes:

    • Region and zone: China (Qingdao), Zone B

    • Instance type: ecs.g5.xlarge

    • Operating system platform: Windows

    • Number of instances: 1

    You have five pay-as-you-go instances with the following configuration:

    • Region and zone: China (Qingdao), Zone B

    • Instance type: ecs.g5.xlarge

    • Image: Windows

    1. View the normalization factor for each instance type. The reserved instance and the pay-as-you-go instance have the same instance type and therefore the same normalization factor. The computing power of ecs.g5.xlarge is 4.

    2. Compare computing power.

      • Reserved instance: one reserved instance delivers computing power of 4 per hour (1 instance * 4).

      • Pay-as-you-go instances: five instances consume computing power of 20 per hour (5 instances * 4).

    3. Offset result: Based on the computing power comparison, the reserved instance offsets 20% of the bills for five pay-as-you-go instances per hour and can randomly offset the bill for one instance.

    Multiple reserved instances offset one instance

    You have two active zonal reserved instances with the following attributes:

    • Region and zone: China (Qingdao), Zone B

    • Instance type: ecs.g5.xlarge

    • Operating system platform: Windows

    • Number of instances: 1

    You have one pay-as-you-go instance with the following configuration:

    • Region and zone: China (Qingdao), Zone B

    • Instance type: ecs.g5.xlarge

    • Image: Windows

    1. View the normalization factor for each instance type. The reserved instance and the pay-as-you-go instance have the same instance type and therefore the same normalization factor. The computing power of ecs.g5.xlarge is 4.

    2. Compare computing power.

      • Reserved instances: two reserved instances deliver computing power of 8 per hour (1 instance * 4 * 2).

      • Pay-as-you-go instance: one instance consumes computing power of 4 per hour (1 instance * 4).

    3. Offset result: Based on the computing power comparison, two reserved instances can offset 100% of the bill for one pay-as-you-go instance per hour, and 50% of the reserved instances remain unused.

    Multiple reserved instances offset multiple instances

    You have five active zonal reserved instances with the following attributes:

    • Region and zone: China (Qingdao), Zone B

    • Instance type: ecs.g5.xlarge

    • Operating system platform: Windows

    • Number of instances: 1

    You have five pay-as-you-go instances with the following configuration:

    • Region and zone: China (Qingdao), Zone B

    • Instance type: ecs.g5.xlarge

    • Image: Windows

    1. View the normalization factor for each instance type. The reserved instance and the pay-as-you-go instance have the same instance type and therefore the same normalization factor. The computing power of ecs.g5.xlarge is 4.

    2. Compare computing power.

      • Reserved instances: five reserved instances deliver computing power of 20 per hour (1 instance * 4 * 5).

      • Pay-as-you-go instances: five instances consume computing power of 20 per hour (5 instances * 4).

    3. Offset result: Based on the computing power comparison, five reserved instances offset 100% of the bills for five pay-as-you-go instances per hour.

    Examples of resource reservation

    Example scenario

    Zonal reserved instance

    Pay-as-you-go instance

    Reservation result

    Resource reservation

    You have one active zonal reserved instance.

    • Region and zone: China (Qingdao), Zone B

    • Instance type: ecs.g5.2xlarge

    • Operating system platform: Linux

    • Number of instances: 10

    You do not have a pay-as-you-go instance.

    • The reserved instance is idle and continues to incur charges.

    • During the validity period, 10 ecs.g5.2xlarge instances are reserved to ensure that pay-as-you-go instances can always be created in China (Qingdao), Zone B.

    Examples of unsuccessful offsets

    Example scenario

    Zonal reserved instance

    Pay-as-you-go instance

    Offset result

    Offset fails

    You have one active zonal reserved instance. The reserved instance has the following attributes:

    • Region and zone: China (Qingdao), Zone B

    • Instance type: ecs.g5.xlarge

    • Operating system platform: Linux

    • Number of instances: 1

    You have one pay-as-you-go instance. The instance has the following configuration:

    • Region and zone: China (Qingdao), Zone B

    • Instance type: ecs.g5.xlarge

    • Operating system platform: Windows

    The reserved instance fails to match the pay-as-you-go instance because their operating system platforms differ. Therefore, the reserved instance cannot offset the bill for the pay-as-you-go instance.

    Offset fails

    You have one active zonal reserved instance. The reserved instance has the following attributes:

    • Region and zone: China (Qingdao), Zone B

    • Instance type: ecs.g5.xlarge

    • Operating system platform: Linux

    • Number of instances: 1

    You have one pay-as-you-go instance. The instance has the following configuration:

    • Region and zone: China (Qingdao), Zone C

    • Instance type: ecs.g5.4xlarge

    • Operating system platform: Linux

    The reserved instance fails to match the pay-as-you-go instance for the following reasons:

    • The reserved instance and the pay-as-you-go instance are not in the same region and zone.

    • The instance sizes of the reserved instance and the pay-as-you-go instance differ.

    Therefore, the reserved instance cannot offset the bill for the pay-as-you-go instance.