How Network Bandwidth Affects MySQL Performance

Network is a fundamental part of database infrastructure, yet most performance benchmarks are run on a local machine with client and server colocated, ignoring the impact of the network.

For products like MySQL Group Replication, the network becomes even more critical, so this article introduces a simple network setup to study its effects.

Installation : Two bare‑metal servers are connected with a dedicated 10 Gb network. The network interface speed is changed to simulate a 1 Gb link using the command ethtool -s eth1 speed 1000 duplex full autoneg off.

A simple benchmark is run with

sysbench oltp_read_only --mysql-ssl=on --mysql-host=172.16.0.1 --tables=20 --table-size=10000000 --mysql-user=sbtest --mysql-password=sbtest --threads=$i --time=300 --report-interval=1 --rand-type=pareto

, varying the thread count from 1 to 2048. The -innodb_buffer_pool_size is set large enough to keep all data in memory, so the workload is CPU‑bound with no I/O overhead. The operating system used is Ubuntu 16.04.

N1 Benchmark – Network Bandwidth : The first experiment compares 1 Gb and 10 Gb networks. Results show that the 1 Gb network is the bottleneck, with throughput reaching about 116 MiB/s (≈928 Mb/s), which is close to the link capacity.

PMM (Percona Monitoring and Management) graphs confirm that network traffic limits throughput.

N2 Benchmark – Protocol Compression : Enabling MySQL protocol compression with --mysql-compression=on improves results on the 1 Gb network, but on the 10 Gb network the additional CPU required for compression halves the throughput.

N3 Benchmark – Network Encryption : SSL encryption on a 1 Gb network incurs roughly a 10 % performance loss on a single thread but still hits the bandwidth ceiling. On the 10 Gb network, SSL scales only up to about 32 threads, after which performance plateaus due to a scalability issue in OpenSSL 1.0 used by MySQL. OpenSSL 1.1.1 provides better scalability, but a special MySQL build linked against it is required.

Conclusion :

Network performance and utilization affect overall application throughput.

Verify whether the network bandwidth is the limiting factor.

If bandwidth is limited, protocol compression can help; otherwise it may worsen performance.

SSL encryption introduces modest loss at low concurrency (≈10 %) and does not scale well for high‑concurrency workloads.

Read the original article at https://www.percona.com/blog/2019/02/19/how-network-bandwidth-affects-mysql-performance/ .