メインコンテンツまでスキップ
バージョン: 3.11
Features and pricing

ScalarDB Benchmarking Tools

This tutorial describes how to run benchmarking tools for ScalarDB. Database benchmarking is helpful for evaluating how databases perform against a set of standards.

Benchmark workloads​

  • TPC-C
  • YCSB (Workloads A, C, and F)
  • Multi-storage YCSB (Workloads C and F)
    • This YCSB variant is for a multi-storage environment that uses ScalarDB.
    • Workers in a multi-storage YCSB execute the same number of read and write operations in two namespaces: ycsb_primary and ycsb_secondary.

Prerequisites​

  • One of the following Java Development Kits (JDKs):
  • Gradle
  • Kelpie
    • Kelpie is a framework for performing end-to-end testing, such as system benchmarking and verification. Get the latest version from Kelpie Releases, and unzip the archive file.
  • A client to run the benchmarking tools
  • A target database
    • For a list of databases that ScalarDB supports, see Databases.
注記

Currently, only JDK 8 can be used when running the benchmarking tools.

Set up the benchmarking tools​

The following sections describe how to set up the benchmarking tools.

Clone the ScalarDB benchmarks repository​

Open Terminal, then clone the ScalarDB benchmarks repository by running the following command:

git clone https://github.com/scalar-labs/scalardb-benchmarks

Then, go to the directory that contains the benchmarking files by running the following command:

cd scalardb-benchmarks

Build the tools​

To build the benchmarking tools, run the following command:

./gradlew shadowJar

Load the schema​

Before loading the initial data, the tables must be defined by using the ScalarDB Schema Loader. You can download the ScalarDB Schema Loader on the ScalarDB Releases page. Select the Schema Loader based on how you access ScalarDB:

  • Using the ScalarDB Core library (Community edition)?: Choose scalardb-schema-loader-<VERSION>.jar for the version of ScalarDB that you're using. Then, save the .jar file in the scalardb-benchmarks root directory.
  • Using ScalarDB Cluster (Enterprise edition)?: Choose scalardb-cluster-schema-loader-<VERSION>-all.jar for the version of ScalarDB Cluster that you're using. Then, save the .jar file in the scalardb-benchmarks root directory.

In addition, you need a properties file for accessing ScalarDB via the Java CRUD interface. For details about configuring the ScalarDB properties file, see ScalarDB Configurations or ScalarDB Cluster Client Configurations.

After applying the schema and configuring the properties file, select a benchmark and follow the instructions to create the tables.

To create tables for TPC-C benchmarking (tpcc-schema.json), run the following command, replacing the contents in the angle brackets as described:

java -jar scalardb-schema-loader-<VERSION>.jar --config <PATH_TO_SCALARDB_PROPERTIES_FILE> -f tpcc-schema.json --coordinator

If you are using ScalarDB Cluster, run the following command instead:

java -jar scalardb-cluster-schema-loader-<VERSION>-all.jar --config <PATH_TO_SCALARDB_PROPERTIES_FILE> -f tpcc-schema.json --coordinator

Prepare a benchmarking configuration file​

To run a benchmark, you must first prepare a benchmarking configuration file. The configuration file requires at least the locations of the workload modules to run and the database configuration.

The following is an example configuration for running the TPC-C benchmark. The ScalarDB properties file specified for config_file should be the properties file that you created as one of the steps in Load the schema.

注記

Alternatively, instead of using the ScalarDB properties file, you can specify each database configuration item in the .toml file. If config_file is specified, all other configurations under [database_config] will be ignored even if they are uncommented.

[modules]
[modules.preprocessor]
name = "com.scalar.db.benchmarks.tpcc.TpccLoader"
path = "./build/libs/scalardb-benchmarks-all.jar"
[modules.processor]
name = "com.scalar.db.benchmarks.tpcc.TpccBench"
path = "./build/libs/scalardb-benchmarks-all.jar"
[modules.postprocessor]
name = "com.scalar.db.benchmarks.tpcc.TpccReporter"
path = "./build/libs/scalardb-benchmarks-all.jar"

[database_config]
config_file = "<PATH_TO_SCALARDB_PROPERTIES_FILE>"
#contact_points = "localhost"
#contact_port = 9042
#username = "cassandra"
#password = "cassandra"
#storage = "cassandra"

You can define parameters to pass to modules in the configuration file. For details, see the sample configuration files below and available parameters in Common parameters:

Run a benchmark​

Select a benchmark, and follow the instructions to run the benchmark.

To run the TPC-C benchmark, run the following command, replacing <PATH_TO_KELPIE> with the path to the Kelpie directory:

/<PATH_TO_KELPIE>/bin/kelpie --config tpcc-benchmark-config.toml

In addition, the following options are available:

  • --only-pre. Only loads the data.
  • --only-process. Only runs the benchmark.
  • --except-pre Runs a job without loading the data.
  • --except-process. Runs a job without running the benchmark.

Common parameters​

NameDescriptionDefault
concurrencyNumber of threads for benchmarking.1
run_for_secDuration of benchmark (in seconds).60
ramp_for_secDuration of ramp-up time before benchmark (in seconds).0

Workload-specific parameters​

Select a benchmark to see its available workload parameters.

NameDescriptionDefault
num_warehousesNumber of warehouses (scale factor) for benchmarking.1
load_concurrencyNumber of threads for loading.1
load_start_warehouseStart ID of loading warehouse. This option can be useful with --skip-item-load when loading large-scale data with multiple clients or adding additional warehouses.1
load_end_warehouseEnd ID of loading warehouse. You can use either --num-warehouses or --end-warehouse to specify the number of loading warehouses.1
skip_item_loadWhether or not to skip loading item table.false
use_table_indexWhether or not to use a generic table-based secondary index instead of ScalarDB's secondary index.false
np_onlyRun benchmark with only new-order and payment transactions (50% each).false
rate_new_orderPercentage of new-order transactions. When specifying this percentage based on your needs, you must specify the percentages for all other rate parameters. In that case, the total of all rate parameters must equal 100 percent.N/A
rate_paymentPercentage of payment transactions. When specifying this percentage based on your needs, you must specify the percentages for all other rate parameters. In that case, the total of all rate parameters must equal 100 percent.N/A
rate_order_statusPercentage of order-status transactions. When specifying this percentage based on your needs, you must specify the percentages for all other rate parameters. In that case, the total of all rate parameters must equal 100 percent.N/A
rate_deliveryPercentage of delivery transactions. When specifying this percentage based on your needs, you must specify the percentages for all other rate parameters. In that case, the total of all rate parameters must equal 100 percent.N/A
rate_stock_levelPercentage of stock-level transactions. When specifying this percentage based on your needs, you must specify the percentages for all other rate parameters. In that case, the total of all rate parameters must equal 100 percent.N/A
backoffSleep time in milliseconds inserted after a transaction is aborted due to a conflict.0

Footnotes​

  1. The default value is false for use_read_modify_write since Workload A doesn't assume that the transaction reads the original record first. However, if you're using Consensus Commit as the transaction manager, you must set use_read_modify_write to true. This is because ScalarDB doesn't allow a blind write for an existing record. ↩