MySQL vs Azure -- Performance Benchmarks¶

Query performance by Azure MySQL Flexible Server tier (Burstable/General Purpose/Memory Optimized), IOPS comparison, connection pooling, replication lag, and backup/restore times.

These benchmarks are representative, not definitive. Actual performance depends on workload characteristics, data volume, indexing strategy, query complexity, and network latency. Self-hosted MySQL performance varies enormously based on hardware, storage subsystem, and tuning. Always run your own benchmarks with your specific workload before making migration decisions.

1. Test environment¶

1.1 Configurations tested¶

1.2 Test database¶

• Size: 20 GB (50 tables, 200M rows total)
• Schema: OLTP workload (users, orders, products, inventory, sessions)
• Indexes: B-tree on PKs, FKs, and common query columns
• Data distribution: Realistic skew (Zipf distribution on popular products)
• Character set: utf8mb4, collation utf8mb4_0900_ai_ci

2. OLTP transaction throughput¶

2.1 sysbench OLTP read-write benchmark¶

Standard sysbench OLTP read-write workload (10 tables, 1M rows each, 128 threads).

Key observations:

• Azure MySQL GP D16ds_v4 approaches bare-metal performance at lower total cost (managed service eliminates DBA overhead)
• Azure MySQL MO E16ds_v4 slightly outperforms GP D16ds_v4 due to larger buffer pool (128 GB vs 64 GB)
• Burstable tier is suitable for dev/test and low-traffic applications but not high-throughput OLTP
• For most web applications (< 1,000 TPS), Azure MySQL GP D4ds_v4 provides adequate performance

2.2 Read-only benchmark (SELECT heavy)¶

Key observations:

• Read-heavy workloads benefit significantly from Memory Optimized tier (larger buffer pool = more data cached in RAM)
• Azure MySQL MO E16ds_v4 matches or exceeds bare-metal performance for read-heavy workloads
• For read-heavy applications, add read replicas to distribute load across multiple servers

2.3 Write-heavy benchmark (INSERT/UPDATE)¶

Key observations:

• Write performance is heavily dependent on IOPS; pre-provisioned IOPS significantly improve write throughput
• Azure MySQL with pre-provisioned IOPS approaches bare-metal NVMe performance
• For write-heavy workloads, budget for pre-provisioned IOPS rather than a larger compute tier

3. IOPS comparison¶

3.1 IOPS by storage size (baseline)¶

3.2 Pre-provisioned IOPS benchmark¶

4. Connection pooling¶

4.1 ProxySQL vs built-in connection management¶

4.2 Connection scaling by tier¶

4.3 Application connection pool configuration¶

# HikariCP (Java) recommended settings
maximumPoolSize: 20 # Start small, increase based on load
minimumIdle: 5
connectionTimeout: 30000 # 30 seconds
idleTimeout: 600000 # 10 minutes
maxLifetime: 1800000 # 30 minutes
leakDetectionThreshold: 60000

# SQLAlchemy (Python) recommended settings
pool_size: 20
max_overflow: 10
pool_timeout: 30
pool_recycle: 1800
pool_pre_ping: true

5. Replication lag¶

5.1 Read replica lag by workload¶

5.2 Zone-redundant HA failover times¶

6. Backup and restore times¶

6.1 Point-in-time restore¶

6.2 Geo-restore (cross-region)¶

6.3 mysqldump/mydumper export times (for comparison)¶

7. Cost-per-transaction analysis¶

Key finding: Azure MySQL Flexible Server delivers 3-4x lower cost per transaction than self-hosted MySQL when DBA labor and infrastructure are included. Even comparing compute costs alone, Azure is competitive with raw VM costs while providing managed HA, backups, and monitoring.

8. Benchmark methodology¶

8.1 Tools used¶

• sysbench 1.0.20: OLTP read-write, OLTP read-only, OLTP write-only benchmarks
• mysqlslap: Built-in MySQL benchmark tool for simple load testing
• fio 3.35: Storage I/O benchmarking (IOPS, throughput, latency)
• Custom application: Python-based OLTP simulator for realistic workload testing

8.2 Running your own benchmarks¶

# Install sysbench
sudo apt-get install sysbench

# Prepare test data
sysbench oltp_read_write \
  --mysql-host=myserver.mysql.database.azure.com \
  --mysql-user=admin --mysql-password=password \
  --mysql-db=sbtest --tables=10 --table-size=1000000 \
  --mysql-ssl=REQUIRED \
  prepare

# Run benchmark (128 threads, 300 seconds)
sysbench oltp_read_write \
  --mysql-host=myserver.mysql.database.azure.com \
  --mysql-user=admin --mysql-password=password \
  --mysql-db=sbtest --tables=10 --table-size=1000000 \
  --threads=128 --time=300 --report-interval=10 \
  --mysql-ssl=REQUIRED \
  run

# Clean up
sysbench oltp_read_write \
  --mysql-host=myserver.mysql.database.azure.com \
  --mysql-user=admin --mysql-password=password \
  --mysql-db=sbtest --tables=10 \
  --mysql-ssl=REQUIRED \
  cleanup

Next: Best Practices | TCO Analysis | Migration Playbook

Maintainers: csa-inabox core team Last updated: 2026-04-30