Sysbench benchmark results

This section presents database performance benchmark results conducted using the Sysbench tool, providing reference information about the processing capabilities of database engines across different resource configurations (flavors).

Benchmark results are for reference only and do not guarantee actual performance.

1. Introduction to Sysbench benchmark

Purpose

Sysbench benchmark is used to:

• Evaluate the OLTP (Read/Write) processing performance of database engines.
• Observe the degree of performance improvement when increasing CPU and RAM resources.
• Provide reference data for users to select appropriate configurations.

Database engines benchmarked

The database engines benchmarked in this document include: PostgreSQL, MySQL, and MariaDB.

Measurement metrics

Benchmark results include the following key metrics:

• Sysbench Read: Total number of read queries executed during the benchmark.
• Sysbench Write: Total number of write queries executed during the benchmark.
• QPS (Queries Per Second): Average number of queries processed per second.
• TPS (Transactions Per Second): Average number of complete transactions per second.

QPS and TPS metrics are used as the basis for evaluating overall system performance. Higher QPS and TPS values indicate better system throughput.

2. Benchmark results by database engine

2.1. PostgreSQL

Test environment:

Parameter	Value
Benchmark tool	Sysbench (OLTP Read/Write)
Number of tables	64
Records per table	1,000,000
Workload type	Read/Write
Number of threads	Configured per flavor
PostgreSQL version	PostgreSQL 17

Results:

Flavor (vCPU/RAM)	Threads	Sysbench Read	Sysbench Write	QPS	TPS
2C4G	64	2,595,600	741,590	6,177.66	308.88
2C8G	64	2,481,276	708,929	5,905.53	295.27
4C8G	64	3,189,018	911,134	7,589.71	379.48
8C16G	64	4,829,286	1,379,738	11,496.20	574.79
8C32G	64	5,679,842	1,622,732	13,519.46	675.94
16C32G	64	6,448,036	1,842,199	15,350.46	767.49
16C64G	64	6,926,948	1,979,031	16,489.02	824.41

2.2. MySQL

Test environment:

Parameter	Value
Benchmark tool	Sysbench (OLTP Read/Write)
Number of tables	64
Records per table	1,000,000
Workload type	Read/Write
Number of threads	Configured per flavor
MySQL version	MySQL 8.0.42

Results:

Flavor (vCPU/RAM)	Threads	Sysbench Read	Sysbench Write	QPS	TPS
4C8G	16	6,814,500	1,947,000	16,224.39	811.22
8C16G	32	9,748,144	2,785,184	23,209.29	1,160.46
8C32G	32	9,423,834	2,692,524	22,430.67	1,121.53
16C32G	64	9,786,238	2,796,068	23,289.48	1,164.47

2.3. MariaDB

Test environment:

Parameter	Value
Benchmark tool	Sysbench (OLTP Read/Write)
Number of tables	64
Records per table	1,000,000
Workload type	Read/Write
Number of threads	Configured per flavor
MariaDB version	MariaDB 8.0.42

Results:

Flavor (vCPU/RAM)	Threads	Sysbench Read	Sysbench Write	QPS	TPS
4C8G	16	10,573,514	2,111,341	25,174.34	1,258.72
8C16G	32	8,923,236	2,094,628	21,245.25	1,062.26
8C32G	32	8,491,182	2,086,388	20,216.52	1,010.83
16C32G	64	10,267,208	2,568,032	24,444.58	1,222.23
16C64G	64	10,789,884	2,719,241	25,688.30	1,284.42

3. Observations & recommendations

• Increasing CPU and RAM generally improves throughput.
• Each database engine has different scaling characteristics.
• At higher configurations, performance gains may slow down depending on workload and system limits.

Important notes:

• Benchmark results depend on workload and are for reference only.
• Actual performance may vary depending on:
  • Application workload characteristics.
  • Database schema and indexes.
  • Read/write ratio.
  • Storage and network configuration.

Users should perform benchmarks with actual workloads before deploying production environments.

Recommendation:

Use benchmark results as reference data when selecting a database engine and resource configuration. To achieve optimal performance, test with your application's actual workload.