Elastic Plan best practices
For subscription Serverless resource groups, Elastic Plan lets you temporarily add elastic CUs on top of reserved CUs to handle peak demand, billed on a pay-as-you-go basis. Use the following scenarios to find the right balance between cost and performance.
Elastic Plan
DataWorks Serverless resource groups offer two billing methods:
-
pay-as-you-go: You are billed for the
CU-hoursyou consume. This method is ideal for workloads with significant fluctuations, short-term testing and validation, or for handling resource contention during peak business hours. -
subscription: You purchase a resource group with a fixed number of compute units (CUs) in advance. This method offers a lower per-unit cost for CUs during the contract period and is suitable for workloads with stable, predictable usage.
With asubscription resource group, provisioning for peak demand wastes resources during off-peak hours, while provisioning for average demand falls short during peaks.Elastic Plan bridges this gap by letting you temporarily add elastic CUs to your reserved CUs during specified time periods, billed at thepay-as-you-go rate.
To handle peak demand, you can add compute capacity in two ways:
-
Option 1 (subscription upgrade): Directly upgrade the reserved CU specification of your resource group to increase compute capacity, billed at the subscription rate.
-
Option 2 (configure an Elastic Plan): Temporarily add elastic CUs during a specified period. The plan automatically releases the resources after the peak. The elastic CUs are billed at the pay-as-you-go rate.
Cost comparison analysis
The following table compares prices for 1 CU, using the China (Shanghai) region as an example.
|
Billing item |
Price |
Description |
|
Pay-as-you-go (elastic CU rate) |
CNY 0.5/CU-hour |
Billed per minute based on actual usage. |
|
Subscription (reserved CU rate) |
CNY 240/CU/month |
Fixed monthly payment. |
The prices are for example purposes only and are based on the China (Shanghai) region. Actual prices may vary by region. For detailed pricing information, see Billing for Serverless resource groups.
Based on a 30-day month, the cost break-even point is:
-
Monthly cost of elastic CUs = CNY 0.5/CU-hour × H hours/day × 30 days
-
Monthly cost of a subscription upgrade = CNY 240/CU/month
-
Break-even point: 0.5 × H × 30 = 240, which gives H = 16 hours/day
Conclusion: An Elastic Plan is more economical if you need additional capacity for less than 16 hours per day. For 16 hours or more, the consistent demand justifies upgrading your subscription to add reserved CUs instead.
The break-even point is calculated based on prices in the China (Shanghai) region. This point may vary by region. Always calculate the break-even point using the actual prices for your region.
Limitations
-
The Elastic Plan feature is available only forsubscription Serverless resource groups.
-
The scaled-out elastic CUs can be used only forData Computing andSynchronization Tasks in Data Integration quotas. They cannot be used for other scenarios, such asDataService Studio,Individual Development Environment, orModel Service.
-
The maximum number of elastic CUs is the lesser offive times your reserved CU count or1,000 CUs.
-
A single resource group supports up to5 active scaling configurations and20 time plans (excluding disabled plans).
Use cases
Use case 1: Nightly peak and idle day
Scenario
A company runs data computing tasks primarily at night (for example, 8 hours from 22:00 to 06:00 the next day), with only light query and development work during the day. It uses a subscription Serverless resource group with 50 reserved CUs, but needs an additional 50 CUs during the nightly batch processing peak.
Solution comparison
Based on the prices in the China (Shanghai) region, the following table compares the monthly costs of two scale-out options, both assuming an increase of 50 CUs.
|
Comparison item |
Option 1: Upgrade to 100 CUs |
Option 2: Elastic Plan (50 CUs) |
|
Monthly cost of additional capacity |
50 CUs × CNY 240/CU/month = CNY 12,000/month |
50 CUs × CNY 0.5/CU-hour × 8 hours × 30 days = CNY 6,000/month |
|
Savings percentage |
- |
50% savings compared to a subscription upgrade |
|
Resource utilization |
The additional 50 CUs would be idle for 16 hours per day on average, resulting in a utilization rate of about 33.3%. |
Scales out on demand at night and automatically releases resources during the day. |
Configuration recommendations
-
Log on to the DataWorks management console. In the left-side navigation pane, click Resource Group to go to the Resource Groups page.
-
Click the target subscription Serverless resource group and then click Scaling Plan in the upper-right corner to go to the Scaling Plan page.
-
On the Scheduled Elasticity tab, click Create Configuration Plan. Set the configuration name to "Nightly Batch Scale-out" and the number of elastic CUs to 50.
-
On the Scheduled Elasticity tab, click Add Schedule. Set the effective period to 22:00 to 06:00 (next day) daily and associate it with the configuration you just created.
An Elastic Plan requires time to schedule and initialize resources. To ensure that an edited plan takes effect on the same day, we recommend submitting it at least 3 hours before its scheduled start time. You must submit it at least 5 minutes before the start time to allow for necessary system processing.
Use case 2: Accelerate data integration for SLA
Scenario
A company needs to incrementally synchronize business data to a MaxCompute data warehouse every day. The SLA requires synchronization to finish within a 4-hour window from 06:00 to 10:00. The company uses a subscription Serverless resource group with 50 reserved CUs, but needs 150 CUs total during this window to meet the deadline.
Solution comparison
Based on the prices in the China (Shanghai) region, the following table compares the monthly costs of two scale-out options, both assuming an increase of 100 CUs.
|
Comparison item |
Option 1: Upgrade to 150 CUs |
Option 2: Elastic Plan (100 CUs) |
|
Monthly cost of additional capacity |
100 CUs × CNY 240/CU/month = CNY 24,000/month |
100 CUs × CNY 0.5/CU-hour × 4 hours × 30 days = CNY 6,000/month |
|
Savings percentage |
- |
75% savings compared to a subscription upgrade |
|
Resource utilization |
The additional 100 CUs would be idle for 20 hours per day on average, resulting in a utilization rate of about 16.7%. |
Scales out only during the synchronization window and automatically releases resources after completion. |
Configuration recommendations
-
Log on to the DataWorks management console. In the left-side navigation pane, click Resource Group to go to the Resource Groups page.
-
Click the target subscription Serverless resource group and then click Scaling Plan in the upper-right corner to go to the Scaling Plan page.
-
On the Scheduled Elasticity tab, click Create Configuration Plan. Set the configuration name to "Data Sync Acceleration" and the number of elastic CUs to 100.
-
On the Scheduled Elasticity tab, click Add Schedule. Set the effective period to 06:00 to 10:00 daily and associate it with the configuration you just created.
The maximum number of elastic CUs is the lesser of five times your reserved CU count or 1,000 CUs. In this scenario, with 50 reserved CUs, the upper limit for elastic CUs is 50 × 5 = 250 CUs. Scaling out by 100 CUs is within this limit.
Use case 3: Handle festival demand spikes
Scenario
An e-commerce company experiences a massive surge in data processing during a major shopping festival (for example, the Singles' Day event from November 1 to November 11). It needs an additional 100 CUs for about 12 hours each day to support real-time data synchronization and computing tasks. Outside this period, the company's 50 reserved CUs are sufficient.
Solution comparison
Based on prices in the China (Shanghai) region, the following table compares the costs of two scale-out options during the festival, both adding 100 CUs. For a fair comparison, Option 1 (subscription upgrade) assumes you manually upgrade at the start and downgrade immediately after, limiting the additional cost to the 11-day period.
|
Comparison item |
Option 1: Upgrade to 150 CUs |
Option 2: Elastic Plan (100 CUs) |
|
Cost of additional capacity |
100 CUs × CNY 240/CU/month × (11 days / 30 days) = CNY 8,800 |
100 CUs × CNY 0.5/CU-hour × 12 hours × 11 days = CNY 6,600 |
|
Savings percentage |
- |
25% savings compared to a subscription upgrade |
|
Resource utilization |
The additional 100 CUs would be idle for 12 hours per day on average, resulting in a utilization rate of about 50%. |
Scales out on demand only during the festival and automatically releases resources afterward. |
Configuration recommendations
-
Log on to the DataWorks management console. In the left-side navigation pane, click Resource Group to go to the Resource Groups page.
-
Click the target subscription Serverless resource group and then click Scaling Plan in the upper-right corner to go to the Scaling Plan page.
-
On the Scheduled Elasticity tab, click Create Configuration Plan. Set the configuration name to "Festival Capacity" and the number of elastic CUs to 100.
-
On the Scheduled Elasticity tab, click Add Schedule. Set the effective date range from November 1 to November 11 and the effective time period to 08:00–20:00 daily. Then, associate this plan with the configuration you just created.
Best practices
Assess resource requirements
-
Distinguish between baseline and peak usage: Use subscription reserved CUs for stable, consistent demand to keep long-term costs low. Use an Elastic Plan for periodic or temporary peaks to scale out on demand.
-
Match usage precisely: Set the number of elastic CUs based on your actual task requirements to avoid over-provisioning. Elastic CUs and reserved CUs form a single resource pool. The system then allocates CUs from this combined pool as needed.
-
Review your plans regularly: Check historical execution records and resource utilization charts periodically to assess actual usage. Adjust scaling configurations and time plans as needed.
Plan schedules
-
Submit plans in advance: Elastic Plans need time to schedule and initialize resources. Submit your plan at least 3 hours before its start time.
-
Allow time for scale-in: During scale-in, the system waits for running tasks to finish rather than forcibly terminating them. If your tasks have long runtimes, extend the plan's end time to avoid a prolonged scale-in period.
-
Avoid overlapping time periods: The effective periods of multiple time plans cannot overlap.
-
Effective period limitations: Effective periods must be at least one hour long, and the start and end times must be set in 10-minute increments.
Related documents
-
Elastic Plan for Serverless resource groups: Covers the Elastic Plan feature, including setup procedures and configuration details.
-
Billing for Serverless resource groups: Covers billing details by region.