Provisioned Capacity Units for Astra DB Serverless
|
The Astra DB PCU management features are available only for public preview to qualified participants. Development is ongoing, and the features and functionality are subject to change. This preview is governed by your Agreement and the DataStax Preview Terms. This preview covers two sets of features:
DataStax grants access to each set of features separately. To inquire about access to these features, contact your DataStax account representative. |
Provisioned Capacity Units (PCUs) are pre-provisioned capacity for Astra DB Serverless databases to support complex, latency-sensitive workloads and accommodate periodic traffic spikes:
Compared to standard, on-demand capacity, PCUs provide the following benefits:
-
Continuous availability: Where on-demand capacity relies on autoscaling in response to demand, PCUs are always available for your workloads. You can reserve minimum required resources that never scale down even when your databases are idle.
-
Predictable costs: Hourly billing based on pre-determined rates means that you can pay a predictable amount each month.
-
Preparation for volatility: In anticipation of daily spikes and seasonal traffic, you can reserve resources required to handle peak demand and ensure consistent performance.
-
Support for complex and sensitive workloads: You can reserve resources required for consistent performance in complex and latency-sensitive workflows.
To reserve PCUs for your databases, you create PCU subscriptions (PCU groups). To ensure appropriate provisioning and manage costs, make sure you understand capacity allocation and plan your PCU groups before you create PCU groups.
PCU subscriptions (PCU groups)
A Provisioned Capacity Unit (PCU) subscription represents a finite amount of capacity that is reserved for one or more Astra DB Serverless databases. It is like a bucket of capacity that is shared by the databases assigned to that bucket.
A typical PCU subscription has some continuously-available capacity, as well as some additional capacity that can be provisioned and deprovisioned based on database activity. For more information about capacity types, see Capacity allocation.
To start a PCU subscription, you create a PCU group. Each PCU group represents one PCU subscription, and you can configure each group according to the needs of the databases in that group.
You don’t need to use PCUs for all of your databases. You can have some databases that use PCUs and some that do not.
PCUs for Astra DB databases don’t apply to Astra Streaming tenants. However, you can contact your DataStax account representative to discuss dedicated cluster pricing for Astra Streaming and Streaming Capacity Units.
PCU group billing
PCUs are a premium feature that requires an Enterprise plan.
PCU billing depends primarily on your PCU group settings, including the capacity allocation, tenancy type, and cache optimization.
|
PCU groups can incur billed charges even if the PCU group has no databases or the databases don’t use all of the reserved capacity or minimum capacity. For more information, see Capacity allocation. |
Actual database activity can also impact PCU billing. For example, traffic spikes can temporarily increase Hourly Capacity Unit (HCU) usage in PCU groups that allow autoscaling.
PCU usage is in addition to your Enterprise plan’s standard, on-demand usage rates. However, databases that are assigned to PCU groups are not billed for reads (RRUs) and writes (WRUs).
For specific metering rates, see Astra DB Serverless PCU pricing. Standard Astra DB SLAs apply to all PCU subscription options.
Subscription renewal and cancellation
For information about subscription renewal and cancelling PCU subscriptions, see Delete or discontinue PCU groups in Astra DB.
Capacity allocation
|
Capacity allocation for PCU groups is complex and highly variable from use case to use case. However, it’s critical that you understand these concepts before you create PCU groups. DataStax recommends the following approach:
|
For billing purposes, PCUs are classified as Reserved Capacity Units (RCUs) or Hourly Capacity Units (HCUs).
A single PCU group can have some capacity that is billed at the RCU rate and some capacity that is billed at the HCU rate. The amount of capacity and the rates are determined by the PCU group’s settings.
Reserved Capacity Units (RCUs)
The Reserved Capacity Unit (RCU) rate applies requires a long-term committed to an amount of reserved capacity.
This rate is designed for your most complex and latency-sensitive workloads that require long-term, continuously-available resources.
When you create a PCU group, the group’s reserved capacity is the amount of capacity billed at the RCU rate.
RCU commitment period
RCUs require a commitment of one year or until your current Astra Cloud Services credits expire, whichever comes first.
If a PCU group includes reserved capacity, then you are billed for that capacity at the RCU rate, even if there are no databases assigned to the PCU group or the databases are inactive.
However, compared to continuously-provisioned Hourly Capacity Units (HCUs), RCUs offer cost savings in exchange for a long-term commitment to a specified amount of capacity.
You can increase a PCU group’s reserved capacity at any time. However, RCU increases are permanent for the duration of the subscription period.
You can only decrease or terminate an RCU commitment during subscription renewal. For more information, see Delete or discontinue PCU groups in Astra DB.
RCU billing and usage
RCU billing begins as soon as you provision an RCU-metered PCU. This occurs when you create a PCU group where the reserved capacity is greater than zero. This also occurs if you increase the reserved capacity when you edit a PCU group.
RCU instance usage is metered at a base hourly rate and billed to your credit balance each month for 12 months or until your credit term ends, whichever comes first.
Hourly Capacity Units (HCUs)
Capacity billed at the Hourly Capacity Unit (HCU) rate is functionally the same as capacity billed at the RCU rate. However, the HCU rate doesn’t require a long-term commitment, and you can change the amount of capacity billed at the HCU rate at any time throughout the subscription period.
The HCU rate applies to any capacity that isn’t covered by a group’s reserved capacity (RCUs). Typically, the HCU rate applies when you need to provision additional capacity to accommodate traffic spikes or other short-term increases in demand. However, it is possible to have continuously available capacity that is billed at the HCU rate.
There are two ways that PCUs can be billed at the HCU rate:
-
Autoscaling up to the maximum capacity: If a group’s maximum capacity is greater than the minimum capacity, Astra DB can provision additional PCUs up to the maximum when demand exceeds the minimum capacity. These additional PCUs are billed at the HCU rate. When demand decreases, the additional capacity is automatically deprovisioned down to the minimum.
-
Flexibility for the minimum capacity: A group’s minimum capacity is always ready and available for use, regardless of current database demand, and no additional scaling is required to use this capacity.
If a group’s minimum capacity is greater than the reserved capacity, then this surplus capacity is billed for continuous usage at the HCU rate. You might do this intentionally, if you don’t want to make a long-term commitment to reserved capacity, or if you want to pre-provision additional capacity in advance of increased demand.
HCU billing and usage
HCUs don’t require a minimum commitment term. Instead, DataStax bills HCU usage based on the number of hours that each HCU is active each month. HCU instance usage is metered at a base hourly rate and billed to your credit balance each month.
HCU activity depends on the PCU group configuration and database resource requirements.
Continuous HCU usage
Depending on your PCU subscription configuration or database requirements, it is possible to have capacity that is billed for continuous usage at the HCU rate.
For example, if a group’s minimum capacity is greater than the reserved capacity, then this surplus capacity is billed for continuous usage at the HCU rate.
For more information, see Reserved, minimum, and maximum capacity and Capacity billing examples.
Committed workloads and flexible workloads
In the context of Astra DB Serverless PCUs, the terms committed capacity workloads and flexible capacity workloads refer to the composition of a PCU subscription:
-
Committed capacity workloads: PCU groups for committed capacity workloads include at least one reserved PCU, which is billed at the RCU rate, and can include additional capacity billed at the HCU rate.
Committed capacity workloads include continuously-provisioned resources, and they can never scale to zero.
Committed capacity workloads are intended for any database in any environment that requires long-term, continuous availability, such as multi-region databases and latency sensitive workloads.
-
Flexible capacity workloads: PCU groups for flexible capacity workloads are billed entirely at the HCU rate, and they have the option to manually scale to zero.
With flexible capacity workloads, you are billed for continuous HCU usage based on the group’s minimum capacity.
While flexible capacity workloads don’t require a commitment to reserved capacity, they don’t offer cost savings for continuous usage that can be realized at the RCU rate.
However, you can park flexible capacity PCU groups to reduce costs when these resources aren’t needed. This can help manage costs, but it doesn’t eliminate all costs. For example, databases in parked PCU groups incur billed charges for storage.
|
You can use committed capacity workloads for databases in any environment, including development, testing, production, and otherwise. The choice between committed and flexible capacity workloads is based on the capacity requirements of your databases. For example, for databases that need continuous availability, the RCU rate available to committed capacity workloads typically provides cost savings over continuous billing at the HCU rate. |
Reserved, minimum, and maximum capacity
When you create or edit a PCU group, you decide how many PCUs are allocated to the group by setting the reserved, minimum, and maximum capacity.
Reserved
The reserved capacity is the number of PCUs that are billed at the RCU rate.
All committed capacity PCU groups have a reserved capacity of one or more. This represents your long-term commitment for the duration of the subscription period. For more information, see Reserved Capacity Units (RCUs).
For flexible capacity PCU groups, the reserved capacity is always zero because flexible capacity workloads are based on the HCU rate.
Minimum
The minimum is the number of PCUs that you want available at all times to support the databases in a PCU group. This capacity is continuously available, and your databases don’t have to wait for autoscaling to use this capacity.
The minimum capacity can be billed at the RCU rate, the HCU rate, or both:
-
In flexible capacity PCU groups, the minimum is always greater than the reserved capacity, and the minimum capacity is entirely billed at the HCU rate.
-
For a committed capacity PCU group, if the minimum is equal to the reserved capacity, then this capacity is billed entirely at the RCU rate.
In this case, the minimum capacity is the same as the reserved capacity; there is no difference between these values, and you aren’t billed twice.
-
For a committed capacity PCU group, if the minimum is greater than the reserved capacity, then both rates apply:
-
The reserved capacity is billed at the RCU rate.
-
After subtracting the reserved capacity, the remaining minimum capacity is billed at the HCU rate.
-
In any PCU group where the minimum is greater than the reserved capacity, Astra DB must provision enough PCUs to meet the minimum. All PCUs required to meet the minimum are billed for continuous usage at either the RCU or HCU rate, depending on the group’s configuration. For more information, see Capacity allocation.
Maximum
The maximum capacity is the maximum number of PCUs that a group can scale up to as needed.
If the maximum is greater than the minimum, then Astra DB can provision additional PCUs up to the maximum when needed. These PCUs are in use only when required to meet demand, and they are billed at the HCU rate while in use. When demand decreases, the additional capacity is automatically deprovisioned back down to the minimum.
If the maximum is equal to the minimum, autoscaling is disabled.
Capacity billing examples
The following examples illustrate basic differences between provisioning and billing for committed capacity and flexible capacity PCU groups.
Billing example for committed capacity workloads (RCU-based PCU groups)
For this PCU subscription billing example, assume the workload type is committed capacity, and the PCU group has 2 Reserved PCUs, 3 Minimum PCUs, and 5 Maximum PCUs.
The difference between the Minimum and Reserved is 1 (3 - 2 = 1).
Therefore, Astra DB must provision 2 RCUs and 1 HCU to meet the Minimum.
The difference between the Maximum and Minimum is 2 (5 - 3 = 2).
Therefore, depending on database resource requirements, Astra DB can provision and deprovision 1 or 2 additional PCUs as needed, and these PCUs are billed at the HCU rate.
Because the Maximum is 5, Astra DB can’t provision more than 2 additional PCUs at any time.
This group’s billed charges include, at minimum, 2 PCUs at the RCU rate, and 1 always-provisioned PCU at the HCU rate. If demand requires scaling up to the maximum, the billed charges could also include a variable amount of usage at the HCU rate for 2 additional maximum PCUs.
Billing example for flexible capacity workloads (HCU-only PCU groups)
For this PCU subscription billing example, assume the workload type is flexible capacity, and the PCU group has 0 Reserved PCUs, 1 Minimum PCU, and 1 Maximum PCU.
The difference between the Minimum and Reserved is 1 (1 - 0 = 1).
Therefore, Astra DB must provision 1 HCU to meet the Minimum.
The difference between the Maximum and Minimum is 0 (1 - 1 = 0).
Therefore, Astra DB can’t provision any additional PCUs for this group because autoscaling is disabled.
This group’s billed charges include 1 always-provisioned PCU at the HCU rate, assuming the group is never parked.
You can park flexible capacity PCU groups to reduce costs when those resources aren’t needed. For committed capacity workloads, balancing the Reserved, Minimum, and Maximum can help you manage costs by minimizing the amount of always-provisioned capacity and allowing Astra DB to provision additional PCUs as needed.
Provision type (infrastructure tenancy)
By default, PCUs exist in a multi-tenant environment. This means that, at the datacenter level, your databases might share physical hardware with other Astra DB organizations.
If your workloads require a greater degree of physical isolation, you can enable dedicated tenancy when you create a PCU group.
Because Astra DB is a DBaaS offering, dedicated tenancy isn’t complete isolation, and some backend processes might use shared hardware. If you need absolute isolation, consider other DataStax offerings that provide you with more control over your infrastructure and deployments.
|
The provision type doesn’t limit the number of databases that you can put in a PCU group. You can put multiple databases in any PCU group, regardless of the provision type. |
Provision type billing
Dedicated tenancy is an additional customization that has billing implications. DataStax recommends this option only for workloads that require it.
Usage of dedicated PCUs is metered at a base hourly rate and charged to your credit balance each month.
Provision type configuration
-
Provision type applies to all capacity allocated to a PCU group. You cannot mix shared and dedicated tenancy in the same PCU group.
-
You can’t change the provision type setting after you create a PCU group.
-
For multi-region databases, each region’s PCU group must have the same provision type. For example, if a database is deployed to
us-east-1andeu-west-1, you must create a PCU group in each region, and both PCU groups must have the same provision type. For more information about configuring PCU groups for multi-region databases, see Plan your PCU groups in Astra DB.
Cache optimization
For each PCU group, you decide whether the group’s cache storage is standard or cache optimized.
A standard PCU has 500 GB of cache storage, and a cache optimized PCU has 2.2 TB of cache storage and more RAM.
Cache optimization can help reduce latency for large working datasets, including vector search operations. Whether you choose standard or cache optimized depends on your latency tolerance and the size of your working dataset.
For optimal performance, a PCU group needs enough total cache storage to hold the entire working dataset in cache.
To calculate a PCU group’s total cache storage, multiply the number of allocated PCUs by the cache storage per PCU.
For example, if a group has an allocated capacity of 2 without cache optimization, then the group’s total cache storage is 1 TB (2 * 500 GB).
Estimate cache requirements
To estimate your cache requirements, you need to know the size of your database’s working dataset. This includes frequently-read data and the associated metadata, such as indexes.
Any data not stored in cache is offloaded to Object Storage, such as S3. If you don’t mind slightly increased latency to fetch data from Object Storage, then you can reduce your estimated cache requirements. However, this allowance doesn’t apply to indexes, which must remain in cache to be performant.
Storage Attached Indexes (SAI) can significantly increase the amount of metadata in cache. For tables, make sure your SAIs only index the data that your application actually needs to be indexed. For collections, all fields are indexed by default, but you can use selective indexing to exclude or include certain fields from your indexes. However, you must set selective indexing during collection creation; collection indexing cannot be changed later.
Knowing the size of your SSTables can also help estimate your cache requirements. For example, other considerations aside, if your SSTables are 1.5 TB, your PCU group must have either one cache optimized PCU or four standard PCUs.
Cache optimization billing
Cache optimization is an additional customization that has billing implications. DataStax recommends this option only for workloads that require it.
Usage of cache optimized PCUs is metered at a base hourly rate and charged to your credit balance each month.
Cache optimization configuration
-
The cache optimization setting applies to all capacity allocated to a PCU group. You cannot mix standard and cache optimized storage in the same PCU group.
-
You can’t change the cache optimization setting after you create a PCU group.
-
For multi-region databases, each region’s PCU group must have the same cache storage setting. For example, if a database is deployed to
us-east-1andeu-west-1, you must create a PCU group in each region, and both PCU groups must have the same cache storage setting. For more information about configuring PCU groups for multi-region databases, see Plan your PCU groups in Astra DB.
