High-traffic WordPress websites frequently suffer from excessive MySQL database load caused by repeated query execution, dynamic post meta lookups, and un-cached WooCommerce session writes. By offloading complex SQL queries to an in-memory Redis object cache using persistent Unix domain sockets and fine-tuned eviction policies, database execution latency drops from 120ms to under 0.8ms per request.

1. Unix Domain Sockets vs TCP Loopback Latency

Most standard Redis setups connect over the local network loopback interface (127.0.0.1:6379). While functional, TCP loopback introduces kernel TCP stack overhead, packet encapsulation, and context switching.

Connecting WordPress to Redis over a Unix Domain Socket (/var/run/redis/redis-server.sock) bypasses network routing entirely, utilizing direct shared memory buffers for a 25–30% latency reduction:

// wp-config.php
define('WP_REDIS_SCHEME', 'unix');
define('WP_REDIS_PATH', '/var/run/redis/redis-server.sock');
define('WP_REDIS_DATABASE', 0);
define('WP_REDIS_TIMEOUT', 1.0);
define('WP_REDIS_READ_TIMEOUT', 1.0);

2. Memory Sizing & Eviction Strategy: allkeys-lru

When Redis reaches its maximum allocated memory limit (maxmemory), its eviction policy determines how new keys are accepted. The default policy (noeviction) throws hard errors when memory is full, breaking WordPress writes.

For high-traffic WordPress hosting, configuring allkeys-lru ensures that the least recently used keys are safely evicted to make room for active database query results without downtime:

# /etc/redis/redis.conf
maxmemory 512mb
maxmemory-policy allkeys-lru
save "" # Disable unnecessary disk persistence on pure cache instances

3. Non-Persistent Groups & WooCommerce Optimization

Certain WordPress data groups (such as transient tokens, cart sessions, and rate-limiting counters) should never be stored indefinitely in memory. Defining non-persistent groups prevents cache bloat and eliminates stale inventory data:

// wp-config.php
define('WP_REDIS_IGNORED_GROUPS', [
    'counts',
    'plugins',
    'themes',
    'wc_session_id'
]);

4. Benchmarked Performance Gains

The impact of Redis object caching on WordPress query throughput is summarized below:

5. Summary

Implementing Redis object caching over Unix domain sockets with an allkeys-lru eviction policy is the single most effective optimization for scaling high-traffic WordPress stores. WinWinHost managed WordPress plans come pre-configured with dedicated NVMe Redis instances out of the box.