sysbench

• YSQL

Overview

sysbench is a popular tool for benchmarking databases like Postgres and MySQL, as well as system capabilities like CPU, memory and I/O. Follow the steps below to run Sysbench against YugabyteDB.

The YugabyteDB version of sysbench is forked from the official version with a few modifications to better reflect YugabyteDB's distributed nature.

Running the benchmark

1. Prerequisites

Install sysbench using the following steps.

$ cd $HOME
$ git clone https://github.com/yugabyte/sysbench.git
$ cd sysbench
$ ./autogen.sh && ./configure --with-pgsql && make -j && sudo make install

Make sure you have the YSQL shell ysqlsh exported to the PATH variable. You can download ysqlsh if you do not have it.

$ export PATH=$PATH:/path/to/ysqlsh

2. Start YugabyteDB

Start your YugabyteDB cluster by following the steps here.

3. Run the benchmark

There is a handy shell script run_sysbench.sh that loads the data and runs the various workloads.

./run_sysbench.sh --ip <ip>

This script runs all the 8 workloads using 64 threads with the number of tables as 10 and the table size as 100k. If you want to run the benchmark with a different count of tables and tablesize:

./run_sysbench.sh --ip <ip> --numtables <number of tables> --tablesize <number of rows in each table>

4. Run individual workloads (optional)

This section outlines instructions in case you need to run workloads individually. Before starting the workload you need to load the data first.

$ sysbench oltp_point_select        \
      --tables=10                   \
      --table-size=100000           \
      --range_key_partitioning=true \
      --db-driver=pgsql             \
      --pgsql-host=127.0.0.1        \
      --pgsql-port=5433             \
      --pgsql-user=yugabyte         \
      --pgsql-db=yugabyte           \
      prepare

Then you can run the workload as follows

$ sysbench oltp_point_select        \
      --tables=10                   \
      --table-size=100000           \
      --range_key_partitioning=true \
      --db-driver=pgsql             \
      --pgsql-host=127.0.0.1        \
      --pgsql-port=5433             \
      --pgsql-user=yugabyte         \
      --pgsql-db=yugabyte           \
      --threads=64                  \
      --time=120                    \
      --warmup-time=120             \
      run

The choice of different workloads are:

• oltp_insert
• oltp_point_select
• oltp_write_only
• oltp_read_only
• oltp_read_write
• oltp_update_index
• oltp_update_non_index
• oltp_delete

Expected results

Setup

When run on a 3-node cluster with each node on a c5.4xlarge AWS instance (16 cores, 32 GB of RAM, and 2 EBS volumes), all belonging to the same AZ with the client VM running in the same AZ, you get the following results:

10 Tables Each with 100k Rows