Simple benchmark demonstrating the efficiency of APC compared to memcached. Even if the memcached instance is running on localhost, it is a separate process, and PHP has to talk with it via a socket (usually TCP, sometime UNIX). This requires executing many system calls, like select(), accept(), read(), and write(), each of which crosses the bou…

apc_vs_memcached.php

<?php
$loops = 1000;
$data = array( 'a' => 1, 'b' => 2, 'c' => 3 );


// Memcached
$cache = new Memcached();
$cache->addServer( "127.0.0.1", 11211 );

$start = microtime( true );
for ( $i = 0; $i < $loops; $i++ ) {
	$cache->set( 'speed-test', $data );
	$cache->get( 'speed-test' );
}
$avg = 1e6 * ( ( microtime( true ) - $start ) / $loops );
printf( "Memcached::set() / Memcached::get(): %.2fμs\n", $avg );


// APC
$start = microtime( true );
for ( $i = 0; $i < $loops; $i++ ) {
	apc_store( 'speed-test', $data );
	apc_fetch( 'speed-test' );
}
$avg = 1e6 * ( ( microtime( true ) - $start ) / $loops );
printf( "apc_store() / apc_fetch(): %.2fμs\n", $avg );

// Sample output:
//
// Memcached::set() / Memcached::get(): 80.23μs
// apc_store() / apc_fetch(): 2.33μs

A few things to be aware of when moving caches from memcached to APC:

• APC has no concept of LRU eviction so when your cache is full and you want to add new things you can't until TTLs expire or something does a manual cleanup.
• HHVM's APC implementation uses an unbounded pool so if you keep sticking data into it you can risk running the entire process out of memory.
• APC is process local cache so it can't be used share data with another host

Ignoring any or all of these differences may make things much worse rather than much better, but if you are aware of these at the application level APC can certainly be much much faster.