PolarDB:TPC-C test instructions

Data scale

For regular instances, the test uses 1,000 warehouses:

The ultra-large instance test scales to 10,000 warehouses to verify PolarDB-X 1.0 horizontal scalability. Three 32-core Elastic Compute Service (ECS) instances serve as stress test machines to avoid the stress test machines becoming a bottleneck.

Instance specifications

Choose your edition before starting. The test procedure is the same for all editions — only the instance specifications differ.

Step 1: Set up the test tool

This test uses BenchmarkSQL V5.0. BenchmarkSQL does not support the MySQL protocol by default, so a pre-compiled package with MySQL support is provided.

1. Download tpcc.tar.gz to the ECS instance and extract it:

   mkdir tpcc
   tar zxvf tpcc.tar.gz -C tpcc

   The compiled package includes the following modifications to standard BenchmarkSQL:

   File	Change
   src/client/jTPCC.java	Added MySQL type
   src/client/jTPCCConnection.java	Added MySQL syntax support via alias
   src/LoadData/LoadData.java	Disabled large transaction mechanism during data loading
   src/LoadData/LoadDataWorker.java	Disabled large transaction mechanism during data loading
   run/funcs.sh	Added MySQL type
   run/runDatabaseBuild.sh	Removed unnecessary phases
   run/runBenchmark.sh	Tuned default JVM parameters
   run/runLoader.sh	Tuned default JVM parameters
   run/sql.common/foreignKeys.sql	Commented out all foreign key creation statements (PolarDB-X 1.0 does not support foreign keys)
   run/sql.common/indexCreates.sql	Commented out primary key statements; retained two index statements
   run/sql.common/indexDrops.sql	Commented out all primary key deletion statements
   run/sql.common/tableCreates.sql	Added primary keys and shard keys

2. Edit run/props.mysql with your connection details and test parameters:

   Parameter	Purpose	Notes
   warehouses	Number of warehouses (data scale)	1,000 for regular; 10,000 for ultra-large
   loadWorkers	Concurrent data loaders	Each 100 workers generates approximately 20,000 TPS; set JVM memory in runLoader.sh accordingly
   terminals	Concurrent clients during the stress test	Set before running the benchmark
   runMins	Test duration (minutes)	Set before running the benchmark

   // --------- env config --------- //
   db=mysql
   driver=com.mysql.jdbc.Driver
   conn=jdbc:mysql://drdsxxxx:3306/tpcc?useSSL=false&useServerPrepStmts=false&useConfigs=maxPerformance&rewriteBatchedStatements=true
   user=tpcc
   password=tpcc
   
   // The number of warehouses.
   warehouses=1000
   // The number of concurrent loadWorkers used to import data.
   // 100 loadWorkers generate approximately 20,000 TPS.
   // JVM memory for 100 loadWorkers: 4 GB. For 500 loadWorkers: 16 GB (edit runLoader.sh).
   loadWorkers=100
   
   // The number of concurrent terminals in the stress test.
   terminals=1000
   // Test duration in minutes.
   runMins=10
   // ---------- default config ------- //
   runTxnsPerTerminal=0
   limitTxnsPerMin=0
   
   terminalWarehouseFixed=true
   
   // Transaction mix — must sum to 100
   newOrderWeight=45
   paymentWeight=43
   orderStatusWeight=4
   deliveryWeight=4
   stockLevelWeight=4
   
   resultDirectory=my_result_%tY-%tm-%td_%tH%tM%tS

   Key parameters to configure:

Step 2: Load test data

cd tpcc/run
nohup ./runDatabaseBuild.sh props.mysql &

Step 3: Run the benchmark

cd tpcc/run
./runBenchmark.sh props.mysql

During the test, BenchmarkSQL logs progress to the console. When the test completes, the output looks like:

08:56:16,844 [Thread-883] INFO   jTPCC : Term-00, Measured tpmC (NewOrders) = 104230.88
08:56:16,844 [Thread-883] INFO   jTPCC : Term-00, Measured tpmTOTAL = 231664.49
08:56:16,844 [Thread-883] INFO   jTPCC : Term-00, Session Start     = 2019-09-19 08:54:16
08:56:16,845 [Thread-883] INFO   jTPCC : Term-00, Session End       = 2019-09-19 08:56:16
08:56:16,845 [Thread-883] INFO   jTPCC : Term-00, Transaction Count = 465440

Step 4: Interpret the results

The tpmC (NewOrders) value is the headline result. It counts how many NewOrder transactions PolarDB-X 1.0 processed per minute during the test window.

tpmTOTAL covers all five transaction types. Because only NewOrder transactions count toward tpmC, tpmTOTAL is always higher than tpmC.

Compare your result against the reference values in the Test results section to evaluate performance relative to each edition's specifications.

Step 5: Clean up

cd tpcc/run
./runDatabaseDestroy.sh props.mysql