Benchmark between eZ Publish 4 and eZ Publish 3.10 with or without a PHP opcode cache

After 2 alphas and one beta release, eZ Publish 4 rc1 has been released last week. I'm trying to upgrade but I'm facing an encoding problem In the meantime, I made a benchmark between eZ Publish 4rc1 with PHP 5.2.5 and eZ Publish 3.10.0 with PHP 4.4.7 on a Debian Etch system using the Dotdeb packages. I also tested the performances of APC, eAccelerator and XCache opcode cache modules in those configurations.


I'm using the recommended configuration for Virtual Host setup of eZ Publish. I wrote this shell script to test performances of eZ Publish.

#! /bin/sh



[ ! -d $DATA_LOG_DIR ] && mkdir -p $DATA_LOG_DIR

for ini in $CONF_DIR/* ; do
    INI_BASE=`basename $ini`
    echo $INI_BASE
    [ -f $DATA_LOG ]  && rm -f $DATA_LOG
    touch $DATA_LOG
    # active extension
    ln -s $ini $PHP_CONFD/$INI_BASE
    /etc/init.d/apache restart > /dev/null  2>&1 
    sleep 2

    # initialize cache
    wget $URL -O /dev/null > /dev/null 2>&1
    sleep 2

    # tests
    for i in `seq 1 $TESTS` ; do
        echo "  Test $i"
        ab -c $CONCURRENCY -n $REQUESTS $URL | grep 'Requests per' | tr -s ' ' | cut -d ' ' -f 4 >> $DATA_LOG
        sleep $PAUSE_TESTS
    sleep $PAUSE_CONF

    rm -f $PHP_CONFD/$INI_BASE
    /etc/init.d/apache restart

I've run this shell script with a PHP4 setup (eZ Publish 3.10) and then with a PHP5 setup (eZ Publish 4.0rc1) sharing the same database. The script uses 4 configurations of PHP (no opcode cache, apc, eaccelerator, xcache), for each it makes 5 series of 100 requests with a concurrency of 2 with ab (Apache Benchmark) and it logs the mean number of requests per second. There are pauses between tests. I've run those tests on two pages of this site on a dedicated test server, the first one is the /blog page with view cache et cache-block enabled and the second one is /man/linux but with no view cache and no cache-block at all in order to see how eZ Publish 4 and 3.10 performs on retrieving its cache or on building a page from scratch. The first one makes 6 SQL queries and uses 2 cache-block and its view cache. The second one, without content related caches, makes about 100 SQL queries an displays about 10 XML blocks and 3 dynamic lists.

Result with a cached page

Benchmark eZ Publish 4 and 3.10 on a cached page

Without an opcode cache on cached page, eZ Publish 4 is about 10% quicker than eZ Publish 3.10 but with an opcode cache system, the difference is about 50% ! It's interesting to note that with PHP4, eAccelerator seems to be the faster yet, about 10% more than APC or XCache but with PHP5 there's almost no difference (more or less 2%).

Benchmark eZ Publish 4 and 3.10 on a page without cache

In this test, without an opcode cache, eZ Publish 4 is 85% faster than eZ Publish 3.10. And with an opcode, eZ Publish 4 using PHP 5.2.5 is about 150% faster than eZ Publish 3.10 !


No doubt, eZ Publish 4 is really faster than eZ Publish 3.10 and even more with an opcode cache.

I think I'm going to test eZ Publish 4 and 3.10 with apache 1.3 and apache 2.2 and perhaps with different configuration of MySQL 5.0, stay tuned :)