How to Optimize Server Performance: Config, Load Analysis & Kernel Tuning

Server Configuration Selection

Servers consist of CPU, memory, disk and network interfaces, so selecting a configuration means choosing CPU cores, RAM size, disk capacity/type, and bandwidth. Because performance depends heavily on the application implementation, there is no one‑size‑fits‑all configuration; testing is essential.

Start with a modest server, run load tests, and use the results to guide the final hardware choice. For example, an order‑processing service tested on a 4‑core CPU, 16 GB RAM, 10 Mbps bandwidth, 50 GB HDD server handled 50 concurrent requests and 300 TPS, with CPU at ~75% and memory, bandwidth, and disk usage below 50%.

From this you can infer that a server with 4‑core CPU (CPU usage < 75%), 8 GB RAM (memory usage up to 100%), and 5 Mbps bandwidth can support the same load. Scaling to 200 concurrent users and 2400 TPS would require either eight such servers or a single 32‑core, 64 GB RAM, 40 Mbps server, subject to verification by testing.

Note: Backend servers and databases must be tuned together to avoid mismatched capacities. The method described may not suit every scenario.

Server Load Analysis

Performance tuning begins with load analysis, focusing on CPU usage, memory usage, disk I/O, and network bandwidth.

CPU Usage

CPU usage reflects how busy the processor is; at 100% some processes wait. In practice, keep CPU usage below 75% to handle traffic spikes; sustained usage above this threshold suggests adding more servers.

Use htop to monitor CPU, memory, and load.

Install and Use htop

yum install htop -y

Run htop to view per‑core usage. When all cores exceed 75%, the server is overloaded.

The screenshot shows a 4‑core server with three cores above 75% and all cores around 85%, indicating high load.

Memory Usage

Memory usage shows how much RAM is occupied. When physical memory reaches 100%, the system swaps to virtual memory, which is much slower. Keep physical memory usage below 80% and avoid using swap.

Use htop to view memory usage.

The example shows 16 GB total memory with ~10 GB used (62% usage) and swap disabled.

Disk I/O

Disk I/O represents read/write activity, often dominated by database operations under high concurrency. Use iostat to monitor.

Install and Use iostat

yum install sysstat -y

# View overall disk read/write every second
iostat -x 1

Key disk metrics are %idle (desired > 70%) and %util (desired < 70%).

Average Load

Average load is the number of active processes over time; it should be less than the number of CPU cores. htop also shows the load average (1‑min, 5‑min, 15‑min). Values above 75% of core count indicate overload.

Network Usage

Network bandwidth must stay below 80% utilization to avoid latency spikes. Use nload to monitor inbound and outbound traffic.

Install and Use nload

yum install nload -y

nload

When current bandwidth approaches the maximum, the link is saturated.

Server Kernel Parameter Tuning

Beyond hardware, kernel parameters must be tuned for high‑concurrency workloads, especially on front‑end, back‑end, and database servers.

Two main groups of parameters are the maximum open files per process and TCP settings.

Maximum Open Files

Edit /etc/security/limits.conf and add:

* soft    nofile  65535
* hard    nofile  65535
* soft    nproc   65535
* hard    nproc   65535

The asterisk applies to all users; the values take effect after reboot.

TCP Settings

Edit /etc/sysctl.conf and include the following (example values):

# Disable SYN flood protection for high‑concurrency
net.ipv4.tcp_syncookies = 0
# Enable reuse of TIME‑WAIT sockets
net.ipv4.tcp_tw_reuse = 1
# Enable fast recycling of TIME‑WAIT sockets
net.ipv4.tcp_tw_recycle = 1
# Reduce FIN‑WAIT‑2 timeout
tcp_fin_timeout = 30
# Keepalive interval (seconds)
tcp_keepalive_time = 1200
# Local port range for outbound connections
ip_local_port_range = 10246 5535
# SYN backlog size
tcp_max_syn_backlog = 65535
# Max TIME_WAIT sockets
tcp_max_tw_buckets = 5000
# Max packets queued per network interface
net.core.netdev_max_backlog = 65535
# Max TCP connections
net.core.somaxconn = 65535
# Receive buffer defaults
net.core.rmem_default = 8388608
net.core.rmem_max = 16777216
# Send buffer defaults
net.core.wmem_default = 8388608
net.core.wmem_max = 16777216
# Disable TCP timestamps
net.ipv4.tcp_timestamps = 0
# Limit orphaned sockets to mitigate DoS
net.ipv4.tcp_max_orphans = 3276800