Performance Tuning¶

This guide covers optimizing VeloServe for maximum throughput and minimum latency.

Quick Wins¶

Enable OPcache — immediate 2-5x PHP improvement
Use SAPI mode — 10-100x faster than CGI
Enable the built-in cache — eliminates PHP for repeated requests
Match workers to CPU cores — optimal thread utilization

Server Tuning¶

[server]
workers = "auto"           # Uses all CPU cores
max_connections = 10000    # Increase for high-traffic sites
keepalive_timeout = 75     # Keep connections alive
request_timeout = 60       # Prevent slow clients from blocking

Worker Count¶

"auto" — uses the number of CPU cores (recommended)
Set a specific number if you want to reserve cores for other processes

PHP Tuning¶

CGI Mode¶

[php]
workers = 16              # Match to CPU cores
memory_limit = "256M"
max_execution_time = 30

SAPI Mode (Maximum Performance)¶

[php]
mode = "embed"
workers = 16              # Concurrent PHP threads
embed_stack_limit = "512M"
ini_settings = [
    "opcache.enable=1",
    "opcache.memory_consumption=256",
    "opcache.max_accelerated_files=20000",
    "opcache.validate_timestamps=0",
    "opcache.interned_strings_buffer=16",
    "opcache.fast_shutdown=1",
    "output_buffering=4096",
    "realpath_cache_size=4096K",
    "realpath_cache_ttl=600"
]

Cache Tuning¶

[cache]
enable = true
storage = "memory"
memory_limit = "1G"      # Larger = more pages cached
default_ttl = 3600
cache_static = true
static_ttl = 86400       # 24 hours for static assets

Cache Hit Rate¶

Monitor your cache performance:

veloserve cache stats

Aim for a hit rate above 90%. If lower:

Increase memory_limit
Review cache exclusion rules
Increase TTL where appropriate

Operating System Tuning¶

File Descriptors¶

Increase the file descriptor limit for high-concurrency:

# /etc/security/limits.conf
* soft nofile 65536
* hard nofile 65536

Or in the systemd service:

[Service]
LimitNOFILE=65536

TCP Tuning¶

# /etc/sysctl.conf
net.core.somaxconn = 65535
net.ipv4.tcp_max_syn_backlog = 65535
net.ipv4.tcp_tw_reuse = 1
net.core.netdev_max_backlog = 65535

Apply with:

sudo sysctl -p

Benchmarking¶

Use wrk or hey to benchmark:

# Install wrk
sudo apt install wrk

# Benchmark static file
wrk -t4 -c100 -d30s http://localhost:8080/

# Benchmark PHP
wrk -t4 -c100 -d30s http://localhost:8080/index.php

# Using hey (Go-based)
hey -n 10000 -c 100 http://localhost:8080/

Comparison Benchmarks¶

Setup	Req/sec	Latency (p50)	Latency (p99)
VeloServe SAPI + cache	~50,000	<1ms	~5ms
VeloServe SAPI no cache	~10,000	~1ms	~10ms
VeloServe CGI + cache	~20,000	<1ms	~5ms
VeloServe CGI no cache	~500	~50ms	~200ms
Nginx + PHP-FPM	~2,000	~10ms	~50ms
Apache + mod_php	~1,500	~15ms	~75ms

Results measured on a 4-core server with a simple WordPress page.