This topic describes FeatureStore billing.
FeatureStore incurs resource fees for data source configuration and model training.
-
FeatureDB data sources incur charges for data writing, reads, and storage. See the table below for billing details.
Billing item
Description
Fees
Write fees
Charged per 10,000 requests. Each 64 KB of request body data counts as one request.
-
Beijing, Shanghai, Hangzhou, and Shenzhen: USD 0.02271 per 10,000 requests
-
China (Hong Kong), Singapore, Silicon Valley, and Frankfurt: USD 0.03986 per 10,000 requests
Read fees
Charged per 10,000 requests. Each 128 KB of returned data counts as one request.
-
Beijing, Shanghai, Hangzhou, and Shenzhen: USD 0.01135 per 10,000 requests
-
China (Hong Kong), Singapore, Silicon Valley, and Frankfurt: USD 0.01994 per 10,000 requests
Storage fees
Disk storage
Charged per GiB per hour.
-
Beijing, Shanghai, Hangzhou, and Shenzhen: USD 0.00032 per GiB/hour
-
China (Hong Kong), Singapore, Silicon Valley, and Frankfurt: USD 0.00026 per GiB/hour
Memory storage
-
Beijing, Shanghai, Hangzhou, and Shenzhen: USD 0.00931 per GiB/hour
-
China (Hong Kong), Singapore, Silicon Valley, and Frankfurt: USD 0.01516 per GiB/hour
-
-
Hologres data sources require purchasing a Hologres instance, which incurs Hologres resource fees. See Hologres billing overview.
-
Installing the FeatureStore Python SDK on DataWorks requires purchasing an exclusive resource group for scheduling, which incurs DataWorks resource fees. See Billing of exclusive resource groups for scheduling: Subscription.
-
FeatureStore uses MaxCompute to run and export TrainSet sample tables, which incurs MaxCompute public resource fees. See MaxCompute billing overview.
NoteMaxCompute charges for compute resources, storage resources, and data uploads and downloads. PAI training tasks typically do not incur charges for data uploads and downloads.
For pay-as-you-go billing, compute resource fees are distributed as follows: algorithm tasks appear under PAI, while other tasks (e.g., SQL) appear under MaxCompute.