Billing, performance metrics, and purchase recommendations for serverless resource groups-阿里云帮助中心

Billing scenarios

Fees for a DataWorks serverless resource group include a resource usage fee and a task scheduling fee.

Resource usage fee: Some DataWorks tasks consume compute units (CUs) from a serverless resource group during execution. The system calculates fees based on the total CUs consumed. This fee is billed in CUs.

Where, 1 CU = 1 CPU core + 4 GiB of memory.
Task scheduling fee: A task scheduling fee is incurred when you publish a task for periodic scheduling in the production environment. These tasks are charged only the task scheduling fee, not a resource usage fee. The task scheduling fee is calculated based on the number of successful instance runs, excluding dry runs.

A serverless resource group supports a maximum of 200 concurrent instances. This limit matches or exceeds the concurrency of all legacy resource group types. You do not need to consider the CU specifications for the serverless resource group.

Task type	Task type description	Cost type
Data Integration	Run a data synchronization task, such as batch synchronization, in Data Integration or Data Studio.	Resource usage fee
Data Compute	Run compute node tasks such as PyODPS, Shell, and EMR Hive in Data Studio. Run compute node tasks such as Hologres SQL and EMR Hive in Data Analysis. Run data quality rules (for example, custom EMR SQL). Important For data compute task types, see Task types and CU consumption.
Data Service	Call API endpoints in Data Service.
Personal development environment	Debug tasks using a personal development environment.
Large model service	Deploy and use large model services.
Task scheduling	Periodic scheduling tasks run in the production environment.	Task scheduling fee

Notes

With pay-as-you-go serverless resource groups, resource contention may occur during peak hours, and timely resource availability cannot be fully guaranteed.
You can convert a pay-as-you-go resource group to a subscription resource group, but subscription resource groups cannot be converted to pay-as-you-go resource groups.
When new users activate DataWorks, a pay-as-you-go serverless resource group is purchased by default. No fees are incurred if the resource group is not used. For billing details, see Billing standards.
The total available CUs for a subscription serverless resource group equals the CU amount specified at purchase. Actual usage does not exceed this limit. To get more CUs, scale up the resource group.
When all CUs in a subscription serverless resource group are occupied, newly submitted tasks enter a queuing state and do not start until CUs become available.

Performance metrics

Serverless resource groups are billed based on CU usage, where 1 CU = 1 CPU core + 4 GiB of memory. Plan the resource group specifications based on your development scenarios and task types.

Important

The following recommended specifications are general guidelines. Adjust resources based on your business requirements to ensure tasks run efficiently and stably.

Data Integration

Batch synchronization

Batch synchronization task concurrency	Recommended specifications	Minimum specifications
<4	0.5 CU	0.5 CU
>=4	`(Concurrency - 4) × 0.07 + 0.5` CU	0.5 CU

Real-time synchronization

Synchronization task type		Recommended specifications	Minimum specifications
MySQL real-time synchronization	1 database	2 CU	Minimum specifications for running a real-time synchronization task: 1 CU
	2 to 5 databases	2 CU
	6 or more databases	2 CU
Kafka real-time synchronization		1 CU
Other single-table real-time tasks		1 CU
Full-database real-time synchronization		-	Minimum specifications for running a full-database synchronization task: 2 CU

Data compute

Each data compute task has a default CU allocation. For details, see Task types and CU consumption.

Data Service

Maximum QPS	Minimum specifications	Service availability (SLA)
500	4 CU	99.95%
1000	8 CU
2000	16 CU

Personal development environment

CPU-based personal development environments provide resource quotas from 2 to 100 CUs. GPU-based personal development environments provide resource quotas from 21 to 60 CUs. Estimate requirements based on task types:

Lightweight tasks (such as simple SQL queries and Python script debugging): A lower resource quota (such as 2 CUs) is recommended.
Moderately complex tasks (such as data processing and notebook analysis): A medium resource quota (such as 4 CUs) is recommended.
Deep learning tasks (such as TensorFlow and PyTorch model training): GPU-type resources are recommended. Select an appropriate GPU memory size and CU count based on the model size.

Large model service

Estimate the required CU deduction based on the GPU memory.

Deploying 0.6B, 1.7B, 4B, or 8B models requires a minimum of 24GB GPU memory.
Deploying a 14B model requires a minimum of 48GB GPU memory.
Deploying a 32B model requires a minimum of 96GB GPU memory.

Task scheduling

A serverless resource group supports a maximum of 200 concurrent instances. You do not need to plan for CU specifications. The default concurrency is 50, and you can increase the maximum task scheduling concurrency to 200 on the resource group details page.

Billing models

Serverless resource groups are available in two billing models: subscription (prepaid) and pay-as-you-go (postpaid).

Serverless resource group (subscription): You estimate the required CU amount and usage duration in advance and pay the corresponding fee. Apart from the subscription fee, no additional resource usage fees are charged by DataWorks for data synchronization, data compute, or debugging and calling Data Service APIs.
Serverless resource group (pay-as-you-go): You use product features first and pay based on actual CU consumption afterward. Running certain tasks (such as batch synchronization, Data Service, and data development tasks) on a pay-as-you-go serverless resource group incurs resource usage fees.

The following table compares the two billing models:

Comparison item	Serverless resource group (pay-as-you-go)	Serverless resource group (subscription)
Total available CUs	Calculated based on actual usage.	The CU amount specified at purchase.
Scale-up, scale-down, and renewal	Not applicable	Supported
Quota management	Controls the maximum CUs available for different scenarios. Supported for Data Compute, Data Integration, Data Service, and personal development environments.
Maximum task scheduling concurrency	Supported. A maximum of 200 task instances can run concurrently.
Number of associated VPCs	Data Compute and Data Integration: A maximum of 2 in total. Data Service: Only 1 can be associated.	Depends on the purchased CU amount. 10 CUs or less: A maximum of 4 VPCs in total. Data Compute: Only 1 can be associated. Task scheduling and Data Integration: A maximum of 3 in total. More than 10 CUs: A maximum of 8 VPCs in total. Data Compute: Only 1 can be associated. Task scheduling and Data Integration: A maximum of 7 in total.

Billing standards

Subscription resource group billing

Billing is based on CU usage: Fee = Monthly unit price × Number of months × CUs purchased per month.

Note

The minimum purchase for subscription is 2 CUs per month. There is no upper limit, but availability may be affected by inventory. Check the purchase page for notifications when inventory is insufficient.
If the specifications do not meet your requirements after purchase, you can scale up at any time. For details, see Scale up a resource group.
For the minimum specifications required by different task types when running on a serverless resource group, see Performance metrics.

Region	Monthly unit price (CNY/month/CU)
China (Zhangjiakou)	180.3046
China (Ulanqab)	214.5624
China (Shanghai), China (Hangzhou), China (Beijing), China (Shenzhen)	240
China (Chengdu)	196.7513
UK (London)	329.5431
US (Virginia)	348.3241
Malaysia (Kuala Lumpur)	409.3401
China (Hong Kong), Singapore, Germany (Frankfurt), Indonesia (Jakarta)	436.7817
US (Silicon Valley)	469.9517
Japan (Tokyo)	500.3647
South Korea (Seoul)	366.04404449
UAE (Dubai)	523.8579
Thailand (Bangkok)	347.01794031

Pay-as-you-go resource group billing

Billing is based on CU-hours × CU unit price: Fee = CU-hours × CU unit price. Bills are generated hourly.

Important

In resource group quota management, if you configure 1 CU for Data Service, CU consumption continues regardless of whether the Data Service feature is in use. The consumption stops only after you set the Data Service quota to 0 CUs.

Region	Unit price (CNY/CU-hour)	Example
China (Zhangjiakou)	0.375635	Example: A data synchronization task in the China (Shanghai) region is configured with 2 CUs and completes in 0.5 hours. The CU unit price for the China (Shanghai) region is CNY 0.5/CU-hour. The CU-hours and fee are as follows: CU-hours: 2 CUs × 0.5 hours = 1 CU-hour Fee: 1 CU-hour × CNY 0.5/CU-hour = CNY 0.5
China (Ulanqab)	0.447005
China (Shanghai), China (Hangzhou), China (Beijing), China (Shenzhen)	0.5
China (Chengdu)	0.409899
UK (London)	0.686548
US (Virginia)	0.725675
Malaysia (Kuala Lumpur)	0.852792
Germany (Frankfurt), Indonesia (Jakarta), China (Hong Kong), Singapore	0.909962
US (Silicon Valley)	0.979066
Japan (Tokyo)	1.042426
South Korea (Seoul)	0.76259176
UAE (Dubai)	1.091371
Thailand (Bangkok)	0.72295404

View billing details

When you view billing details in the Billing Management console, the billing items and billing codes for serverless resource groups are:

Pay-as-you-go: The billing item is General Resource Group CU-Hours (Pay-As-You-Go), and the billing code is exresource_cu_hour_post.
Subscription: The billing item is General Exclusive Resource Group Subscription (Hybrid Billing), and the billing code is cu_number.

For detailed instructions, see View billing details.

Expiration and renewal

If your subscription serverless resource group is about to expire, you can renew it. If you do not renew, the resource group is suspended or released. For renewal details, see Renew a resource group.

Scale-up and scale-down fees

Subscription serverless resource groups support scale-up and scale-down after purchase. For the fee calculation logic, see Scale-up and scale-down fees.

Next step

You can purchase a resource group and use it for Data Integration, data development, and Data Service tasks. For information about how to purchase a resource group, associate it with a workspace, and configure network connectivity, see Create and manage a serverless resource group.

Additional information

Appendix 1: Task types and CU consumption

Tasks generated from DataWorks node development are classified into data compute tasks (which consume CUs) and scheduling tasks (which do not consume CUs).

Determine the task type

You can go to the editing page of a node in Data Studio and check the task type in the right-side navigation bar under Schedule Settings > Scheduling Policy.

Compute tasks: In the Scheduling Policy section, you must specify the compute CUs required for the task.
- Scenario 1: The compute CU value can be customized.
  
  In the right-side panel, click Schedule Settings, select the Scheduling Policy tab, and configure the Resource Group and Compute CU (for example, 0.25).
- Scenario 2: The compute CU value can only be set to the default.
  
  In the right-side panel, click Schedule Settings, select the Scheduling Policy tab. The Compute CU default value is 0.25, and the interface displays the message "The current node uses the default CU value. No CU modification is needed."
Scheduling tasks: In the Scheduling Policy section, you can only select a scheduling resource group. No CU configuration is required.

Compute task CU configuration list

Running data compute tasks on a serverless resource group consumes CUs. The default CU and running CU are described as follows:

Default CU: The recommended CU amount allocated by the platform for each task run based on the task type. Running a task below this value may not guarantee efficient execution.
Running CU: The actual CU amount configured for a task run. The platform pre-fills this with the default CU value, which you can adjust as needed. Configuration principles:
- The minimum configuration is 0.25 CU, with a step size of 0.25 CU. If the interface displays The CU Quota of the Current Resource Group Is Insufficient, you can adjust the CU quota for data compute tasks.
- To avoid insufficient or excessive resource allocation, configure resources based on the default CU and the CU quota for data compute tasks. For details, see Configure the CU quota.

Note

Only some tasks support adjusting the running CU. For example:

The running CU for Hologres SQL tasks cannot be adjusted and can only be set to 0.25 (the default CU).
The running CU for PyODPS 2 tasks defaults to 0.5 and can be adjusted as needed (for example, 0.4 or 0.6).

Node type	Node name	Default CU (unit: CU)	Running CU adjustable
Notebook	Basic notebook development	0.5	Yes
MaxCompute	PyODPS 2 node	0.5	Yes
	PyODPS 3 node	0.5	Yes
	MaxCompute MR node	0.5	Yes
	Map metadata to Hologres	0.25	Yes
	Synchronize data to Hologres	0.25	Yes
Hologres	Hologres SQL node	0.25	-
	Synchronize data to MaxCompute	0.25	-
	Maxcompute schema sync node	0.25	Yes
	One-click MaxCompute data synchronization node	0.25	Yes
EMR	EMR Hive node	0.25	-
	EMR Impala node	0.25	-
	EMR MR node	0.25	Yes
	EMR Presto node	0.25	-
	EMR Shell node	0.25	Yes
	EMR Spark node	0.5	Yes
	EMR Spark SQL node	0.5	Yes
	EMR Spark Streaming node	0.5	Yes
	EMR Trino node	0.25	-
	EMR Kyuubi node	0.25	-
Serverless Spark	Serverless Spark Batch node	0.25	-
	Serverless Spark SQL node	0.25	-
	Serverless Kyuubi node	0.25	-
Severless StarRocks	Serverless StarRocks SQL node	0.25	-
Large model	Large language model node	0.5	-
ADB	AnalyticDB for PostgreSQL node	0.25	Yes
	AnalyticDB for MySQL node	0.25	Yes
	ADB Spark node	0.25	-
	ADB Spark SQL node	0.25	-
CDH	CDH Hive node	0.25	-
	CDH Spark node	0.5	Yes
	CDH Spark SQL node	0.25	-
	CDH MR node	0.25	-
	CDH Presto node	0.25	-
	CDH Impala node	0.25	-
Lindorm	Lindorm Spark node	0.25	-
Lindorm	Lindorm Spark SQL node	0.25	-
Click House	ClickHouse SQL	0.25	-
Data Quality	Quality monitoring	0.25	-
Data Quality	Data comparison	0.5	Yes
General	Assignment node	0.25	Yes
	Shell node	0.25	Yes
	OSS object inspection node	0.25	-
	Python node	0.5	Yes
	for-each node	0.25	Yes
	Do-while node	0.25	Yes
	Function Compute node	0.25	-
	SSH node	0.25	-
	Data push node	0.25	-
Database nodes	MySQL node	0.25	-
	SQL Server
	Oracle node
	StarRocks node
	DRDS node
	PolarDB MySQL node
	PolarDB PostgreSQL node
	MariaDB node
	Redshift node
	Saphana node
	Vertica node
	DM (Dameng) node
	KingbaseES node
	OceanBase node
	DB2 node
	GBase 8a node
Algorithm	PAI DLC node	0.25	-

Scheduling task configuration list

Scheduling tasks do not consume CUs from the serverless resource group.

Node type	Node name
Data Integration	Batch synchronization node
Data Integration	Real-time synchronization node
MaxCompute	MaxCompute SQL node
	SQL script template node
	MaxCompute Script node
	MaxCompute Spark node
Flink	Flink SQL Streaming node
Flink	Flink SQL batch node
General	Zero load node
	Parameter node
	Merge node
	Branch node
	Check node
	HTTP trigger node
Algorithm	PAI Designer nodes
Database nodes	PostgreSQL node
	Doris node
	SelectDB node

Appendix 2: Billing modes for task execution

When you run a node task in DataWorks, the compute fee is not necessarily charged by DataWorks. Identify which compute engine or resource the task ultimately runs on. The following three scenarios apply:

Note

A task scheduling fee is always incurred when a task is published to the production environment for periodic scheduling.

Execution mode	Representative task nodes	Compute resource provider	Fee composition
Mode 1: Compute tasks are submitted to the serverless resource group for execution	PyODPS, Shell, Data Integration, Data Quality	Serverless resource group	Serverless resource group fee only
Mode 2: Compute tasks are submitted to third-party engines through the serverless resource group	EMR Hive, Hologres SQL	Serverless resource group + third-party engine	Serverless resource group fee + third-party engine fee
Mode 3: Scheduling tasks are submitted to third-party engines through Operation Center	MaxCompute SQL, Flink SQL	Third-party engine	Third-party engine fee

Appendix 3: Fee breakdown by module

When using a serverless resource group with the following modules, the serverless resource group fees are:

Data Integration: Data Integration tasks run in the Data Integration, Data Studio, and Operation Center modules, consuming the serverless resource group and incurring Data Integration fees. Periodic synchronization tasks also incur task instance scheduling fees.
Data Studio: Data compute tasks and scheduling tasks run in the Data Studio, Data Quality, and Operation Center modules, consuming the serverless resource group and incurring data compute fees and task instance scheduling fees. Using a personal development environment also incurs personal development environment fees. Using large model services or large model nodes also incurs large model service fees.
Data Analysis: SQL query analysis and query result downloads in Data Analysis consume the serverless resource group and incur data compute fees. Using data insights also incurs task instance scheduling fees.
Data Service: You configure the occupied CUs for Data Service through resource group quota management, consuming the serverless resource group and incurring Data Service fees. Using Data Push also incurs task instance scheduling fees.