This topic describes the billing unit of Realtime Compute for Apache Flink.

The basic billing unit of Realtime Compute for Apache Flink is compute unit (CU), which indicates computing resources. One CU is equal to 1 CPU core and 4 GB of memory. The number of CUs determines the computing capability of the underlying system of Realtime Compute for Apache Flink.

The number of CUs that is consumed by a Realtime Compute for Apache Flink job varies based on the queries per second (QPS) of the input data stream of the job, the computing complexity, and the distribution of the input data. To estimate the processing capability of one CU in Realtime Compute for Apache Flink, select one of the following methods based on the operation complexity:
  • For simple operations such as single-stream filtering and string conversion, one CU can process 10,000 data records per second.
  • For complex operations such as operations that use a JOIN clause, GROUP BY clause, or window function, one CU can process 1,000 to 5,000 data records per second.
You can estimate the number of CUs that you need to purchase based on your business scale and the preceding computing capability.
Note
  • The preceding computing capability estimate refers only to the internal processing capability of Realtime Compute for Apache Flink. The external data read and write capabilities are not included. The external data read and write efficiency may affect the estimation of the computing capability of Realtime Compute for Apache Flink.
    • If you want to use Realtime Compute for Apache Flink to read data from Log Service but the query quota of Log Service is limited, the overall computing capability of Realtime Compute for Apache Flink is subject to the capability allowed by Log Service.
    • If the number of connections or transactions per second (TPS) is limited for the ApsaraDB RDS database that Realtime Compute for Apache Flink references, the throughput of Realtime Compute for Apache Flink is limited by the throttling of the ApsaraDB RDS database.
  • If you use window functions in a Realtime Compute for Apache Flink job, the number of CUs consumed in the job is greater than that consumed in a simple job. We recommend that you purchase at least four CUs for such a job.