All Products
Search

This Product

    Realtime Compute for Apache Flink:Performance white paper: Nexmark benchmark

    Document Center

    Realtime Compute for Apache Flink:Performance white paper: Nexmark benchmark

    Last Updated:Mar 30, 2026

    This document describes how to use the Nexmark benchmark suite to evaluate the stream processing performance of Realtime Compute for Apache Flink.

    Performance at a glance

    Nexmark is an industry-standard benchmark suite for stream processing engines. It includes 21 queries that cover typical scenarios such as filtering, aggregation, JOINs, and windows. This document presents a comprehensive performance evaluation of Realtime Compute for Apache Flink using the Nexmark benchmark suite, an 8 CU configuration, and a baseline of 100 million input records. The results show:

    • The RPS for simple queries, such as q0, q1, and q2, can reach 4 to 6.5 million.

    • The RPS for complex aggregation and window queries, such as q4, q5, and q16, ranges from 150,000 to 630,000.

    Test tool

    Nexmark is a standard performance benchmark suite for stream processing engines. The test model is as follows:

    • Nexmark source table: Generates test data (Person, Auction, and Bid events) at a specified rate.

    • Transformations: 21 standard Nexmark queries that cover typical scenarios such as filtering, transformation, aggregation, JOINs, and windows.

    • Blackhole result table: Writes data to a blackhole sink to eliminate performance interference from external storage, focusing the evaluation on the processing capabilities of the Flink engine.

    The Nexmark test tool used in this document is built on the OpenAPI of Realtime Compute for Apache Flink. It automates the full workflow, including job creation, deployment, runtime monitoring, and result collection. You do not need to manually write SQL or create jobs in the console.

    Test environment

    The Flink jobs in this test used the following optimization configurations:

    Parameter

    Value

    Description

    table.exec.mini-batch.enabled

    true

    Enables Mini-Batch aggregation.

    table.exec.mini-batch.allow-latency

    2s

    The buffering interval for Mini-Batch.

    table.optimizer.distinct-agg.split.enabled

    true

    Enables split optimization for Distinct aggregation.

    execution.checkpointing.interval

    3min

    The checkpoint interval.

    Prerequisites

    Procedure

    Step 1: Download the test tool

    Download and extract the Nexmark test tool package nexmark-flink.tar.gz.

    After extraction, the directory structure is as follows:

    nexmark-flink/
├── run_nexmark.sh          # Test entry script
├── nexmark_env.sh          # Environment variable configuration file (requires editing)
├── bin/                    # Runtime scripts
├── conf/                   # Flink job configurations
├── lib/                    # JAR package (to be uploaded to the console)
└── queries-vvp/            # Nexmark Query SQL files

    Step 2: Upload the Nexmark JAR

    1. Log on to the Realtime Compute for Apache Flink console.

    2. Click to enter the target namespace, and in the left navigation bar, select File Management > Upload Resource.

    3. Select and upload the nexmark-flink-0.2-SNAPSHOT.jar file. This file is in the nexmark-flink/lib directory of the test tool.

    4. After the upload is complete, click the file name to copy its OSS path. You will need this path for a parameter in a later step. The file path format varies by storage type:

      • OSS Bucket Storage: oss://<your-bucket-name>/artifacts/namespaces/<your-namespace>/<your-file-name>

        Example: oss://oss-test/artifacts/namespaces/flink-default/nexmark-flink-0.2-SNAPSHOT.jar

      • Fully Managed Storage: oss://flink-fullymanaged-<your-workspace-id>/artifacts/namespaces/<your-namespace>/<your-file-name>

        Example: oss://flink-fullymanaged-e6a123456789/artifacts/namespaces/flink-default/nexmark-flink-0.2-SNAPSHOT.jar

      To view the storage type of your workspace, go to the Realtime Compute for Apache Flink console, find the target workspace, and click Details in the Actions column.

    Step 3: Configure runtime parameters

    Edit the nexmark-flink/nexmark_env.sh file and configure the following parameters.

    Parameter

    Description

    Example

    END_POINT

    The service endpoint for Realtime Compute for Apache Flink. Select an endpoint based on your region. For more information, see Service endpoints.

    ververica.cn-hangzhou.aliyuncs.com

    AK

    The AccessKey ID of your Alibaba Cloud account.

    -

    SK

    The AccessKey Secret of your Alibaba Cloud account.

    -

    WORK_SPACE

    Your workspace ID.

    e6a123456789

    NAMESPACE

    Your workspace namespace.

    flink-default

    NE*******AR

    The OSS path of the JAR package that you uploaded in Step 2.

    oss://flink-fullymanaged-e6a123456789/artifacts/namespaces/flink-default/nexmark-flink-0.2-SNAPSHOT.jar

    FLINK_VERSION

    The Flink engine version to test.

    vvr-11.6-jdk11-flink-1.20

    QUERIES

    The queries to run. Separate multiple queries with commas, for example, q0 or q1,q2,q3. To run all queries, set this parameter to all.

    all

    Note

    Running all queries can be time-consuming. Each query involves stages such as job creation, data generation, and computation. We recommend that you first run a single query (for example, by setting QUERIES to q0) to verify that the environment and parameters are configured correctly before you run the full test suite.

    Step 4: Run the test

    1. In the nexmark-flink directory, run the following command.

      ./run_nexmark.sh

    2. The test tool automatically creates and runs the Nexmark jobs using OpenAPI.

    3. After the test is complete, the output displays the duration of each query in milliseconds. Sample output:

      INFO  com.github.nexmark.flink.vvp.Nexmark - q0 13078
============================================================================
✓ Benchmark execution completed successfully
============================================================================

    Performance results

    The following table shows the performance results of the 21 Nexmark queries on Realtime Compute for Apache Flink with an 8 CU configuration. Each query processed 100 million input records. RPS is calculated by dividing the number of input records by the duration.

    Note

    The following data was collected in a specific test environment. Actual performance may vary with hardware and engine updates. The results are for reference only.

    Query

    Duration (ms)

    RPS

    q0

    23450

    4,264,392

    q1

    22824

    4,381,353

    q2

    15224

    6,568,576

    q3

    21558

    4,638,649

    q4

    157117

    636,468

    q5

    679277

    147,215

    q7

    333837

    299,547

    q8

    29939

    3,340,125

    q9

    266563

    375,146

    q10

    51202

    1,953,049

    q11

    145983

    685,011

    q12

    36991

    2,703,360

    q14

    20012

    4,997,002

    q15

    42734

    2,340,057

    q16

    337293

    296,478

    q17

    27076

    3,693,308

    q18

    96335

    1,038,044

    q19

    95121

    1,051,293

    q20

    231482

    431,999

    q21

    39693

    2,519,336

    q22

    31228

    3,202,254