Why CentOS 7 Boosts MySQL Performance: Real‑World Sysbench Test Results

1 Introduction

Plan to upgrade the database server OS from CentOS 6 to CentOS 7; before upgrading, a performance benchmark is performed. This article shares the benchmark records and results.

2 Benchmark Preparation

2.1 Benchmark Method

Use sysbench‑0.5 to create 10 tables, each with 5 million rows, with concurrency levels 12, 24, 36, 48, 60, 72, each running for 4 hours, totaling 24 hours. The benchmark command is:

/opt/yz-sysbench/bin/sysbench --test=/opt/yz-sysbench/share/sysbench/oltp.lua --oltp-tables-count=10 --oltp-table-size=5000000 --mysql-db=sysbench --mysql-user=sysbench --mysql-password=sysbench --mysql-socket=/srv/my3306/run/mysql.sock --max-time=14400 --max-requests=0 --num-threads=12 --oltp-test-mode=complex run

Note: sysbench is installed locally, so network latency is negligible.

2.2 Target Metrics

Monitor database QPS, TPS, host I/O, CPU, and other performance indicators.

3 Benchmark Data

3.1 QPS

3.2 Insert/Delete

In sysbench OLTP mode, insert and delete ratios are equal, so TPS data is combined.

3.3 Update Performance

3.4 CPU & I/O Comparison

CentOS 7 shows slightly lower CPU load and higher I/O utilization than CentOS 6.

3.5 Flash‑sale Scenario

Refer to the article “Hot‑product Update Optimization Plan” for background.

MySQL uses two parameters to control data flushing. The current setting forces a disk write on every transaction commit, ensuring data safety but causing heavy I/O.

Concurrent updates on the same row can cause lock contention and dead‑lock detection, leading to O(n) complexity for dead‑lock checks, which becomes costly at high concurrency.

Core Optimization Parameters

sync_binlog=0                    -- let the OS flush binlog
innodb_flush_log_at_trx_commit=0 -- flush redo log to disk each second
innodb_deadlock_detect=OFF       -- disable deadlock detection
innodb_lock_wait_timeout=2

Benchmark Scenario

CREATE TABLE `seckill` (
  `id` int(11) DEFAULT NULL,
  `num` bigint(20) DEFAULT NULL
) ENGINE=InnoDB DEFAULT CHARSET=utf8mb4;

INSERT INTO seckill (id, num) VALUES (1, 200000000000000);

mysqlslap -uroot -h127.0.0.1 -P3316 --concurrency=72 --create-schema='test' --query='begin;update seckill set num=num-1 where id =1 and num>1;commit;' --number-of-queries=500000

The test simulates high‑concurrency database load with active sessions of 72, 96, 144, 192, and 256.

In the flash‑sale scenario, CentOS 7 delivers more than three times the performance of CentOS 6 under the same conditions; after tuning, CentOS 7 can handle over 10 k concurrent updates.

3.4 Analysis

QPS on CentOS 7 improves by 20%–40%, and TPS also gains over 20% compared with CentOS 6. Real‑world workloads with complex SQL may see smaller gains.

4 Conclusion

Performance testing of the new system is insightful; CentOS 7 offers several MySQL optimizations. Future posts will explore these in more depth.