OpenMLDB Benchmark

OpenMLDB Benchmak tool is used for tesing the performance of OpenMLDB's online SQL engine.

You may also refer to FEBench (https://github.com/decis-bench/febench), which is a more comprehensive benchmark for real-time feature extraction developed by a third-party, comparing the performance between OpenMLDB and Flink.

Requirements

• CentOS 7 / macOS >= 10.15
• JDK 1.8

Run

1. Compile

   cd benchmark
   mvn clean package

2. Copy the configuration and package

   mkdir -p /work/benchmark/conf /work/benchmark/lib
   cp target/openmldb-benchmark-0.5.0.jar  /work/benchmark/lib
   cp src/main/resources/conf.properties /work/benchmark/conf

3. Modify the configuration

   ZK_CLUSTER=127.0.0.1:32200
   ZK_PATH=/udf_test
   

4. Run benchmark

   cd /work/benchmark
   java -cp conf/:lib/* com._4paradigm.openmldb.benchmark.OpenMLDBPerfBenchmark
   

The above testing run with the default confguration. You can modify WINDOW_NUM, WINDOW_SIZE and JOIN_NUM in the confguration file if you want to evaluate the performance impact of those parameters.

Moreover, the default number of threads is 10. You need to set the thread number by Threads annotation in OpenMLDBPerfBenchmark.java or OpenMLDBLongWindowBenchmark.java as below and compile again if you want to test under other thread confguration.

@Threads(10)

If you want to test Throughput, set BenchmarkMode and OutputTimeUnit annotation in OpenMLDBPerfBenchmark.java or OpenMLDBLongWindowBenchmark.java file as below:

@BenchmarkMode(Mode.Throughput)
@OutputTimeUnit(TimeUnit.SECONDS)

Long Window Benchmark (The Pre-Aggregate Optimization)

Update WINDOW_SIZE in confguration file and execute the following command.

java -cp conf/:lib/ com._4paradigm.openmldb.benchmark.OpenMLDBLongWindowBenchmark

Note: If you want to test specific SQL, you can modify here

Memory Usage Benchmark

Compare with Redis

We provide two ways to measure OpenMLDB's memory usage. First, you can conduct comparative testing using a dataset composed of randomly generated data in a specified format. Second, you can conduct comparative testing using the existing TalkingData dataset.

1. Configure memory.properties

   You need to configure the service addresses for your OpenMLDB and Redis instances in the configuration file. Additionally, you can configure parameters such as KEY_LENGTH, VALUE_LENGTH, VALUE_PER_KEY, and TOTAL_KEY_NUM based on your requirements, which are used for Approach One.

2. Start the test by executing the following command:

   # using generated dataset
   java -cp conf/:lib/* com._4paradigm.openmldb.memoryusagecompare.BenchmarkMemoryUsage
   
   # using TalkingData
   java -cp conf/:lib/* com._4paradigm.openmldb.memoryusagecompare.BenchmarkMemoryUsageByTalkingData
   

3. The test report will be printed at the end of the test execution.

Index Memory Usage

Similar to memory usage comparison testing, we also provide two approaches for testing index memory usage.

1. Configure memory.properties.

   Configure the service addresses for your OpenMLDB.

2. Start the test by executing the following command:

   # using generated dataset
   java -cp conf/:lib/* com._4paradigm.openmldb.memoryusagecompare.BenchmarkIndexMemoryUsage
   
   # using TalkingData
   java -cp conf/:lib/* com._4paradigm.openmldb.memoryusagecompare.BenchmarkIndexMemoryUsageByTalkingData
   

3. The test report will be printed at the end of the test execution.

Note:

1. You can get the TalkingData sample dataset from here, or you can get the full TalkingData training dataset from here.
2. The full TalkingData dataset is too large, so we recommend extracting a portion of it for testing.

Query Performance Comparison

Comparing the query performance of OpenMLDB and Redis under the same scenario.

1. Configure memory.properties.

   Configure the service addresses for your OpenMLDB and Redis instances, and modify the configuration parameters in the 'query perf test' section according to the requirements.

2. (Optional) Load data into OpenMLDB.

3. Start the test by executing the following command:

   java -cp conf/:lib/* com._4paradigm.openmldb.memoryusagecompare.BenchmarkQueryPerf
   

4. The test report will be printed at the end of the test execution.