Yves
(Administrator)
Show profile
Link to this post
Subject: Quick & dirty performance test
Hi,
here is the results of my short performance test I just made. It was primarily meant to get an impression of the some PHP variants I have installed on my test server. But this may also be of public interest, so I simply published it here. The absolute performance values may of course vary on other systems.
Note: This can only cover the overall performance, no detailed information about database utilisation and thus how much faster UNB 1.7 can possibly be.
Scenario:
Software: Debian Linux 3.1, Kernel 2.6.14, Apache 2.0.55, MySQL 5.0.17-max, VMWare 5 inside Windows XP
Hardware: Athlon 64 3000+, 192MB guest RAM, 2xSATA-RAID0 disk on host
Clean install of UNB.devel.20051222
Created 3 empty forums, logged in as admin, benchmarked overview page
Test:
ab -n 50 -c 10 $URL
(ab is a benchmark tool provided with Apache2)
Ran some iterations until results became stable enough.
Results:
PHP4-CGI: 4,2#/sec at avg. 240ms (100%)
PHP4-FastCGI: 5,6#/sec at avg. 177ms (133%)
PHP5-FastCGI: 5,0#/sec at avg. 200ms (119%)
PHP5-FastCGI+APC: 14,8#/sec at avg. 68ms (352%)
Interpretation:
PHP5 is indeed slower than PHP4.
FastCGI seems faster than CGI, even for complex PHP applications.
APC can speed up UNB to 3x.
Maybe some background on the PHP installation:
PHP4-CGI is PHP 4.4.1 invoked through suPHP to get the script running in the correct user context. Configuration changes: register_globals, register_argc_argv. No further extensions.
PHP4-FastCGI is the same PHP, only invoked through Apache's suEXEC and mod_fastcgi.
PHP5-FastCGI is PHP 5.1.1 invoked through the same mod_fastcgi. Configuration changes: register_argc_argv. No further extensions.
PHP5-FastCGI+APC is the same PHP5-FastCGI, only with the APC extension loaded and active, with the APC optimiser disabled. This caches the compiled intermediate code inside PHP and thus saves parsing the large codebase on every request. This is the most impressive effect of the whole test.
here is the results of my short performance test I just made. It was primarily meant to get an impression of the some PHP variants I have installed on my test server. But this may also be of public interest, so I simply published it here. The absolute performance values may of course vary on other systems.
Note: This can only cover the overall performance, no detailed information about database utilisation and thus how much faster UNB 1.7 can possibly be.
Scenario:
Software: Debian Linux 3.1, Kernel 2.6.14, Apache 2.0.55, MySQL 5.0.17-max, VMWare 5 inside Windows XP
Hardware: Athlon 64 3000+, 192MB guest RAM, 2xSATA-RAID0 disk on host
Clean install of UNB.devel.20051222
Created 3 empty forums, logged in as admin, benchmarked overview page
Test:
ab -n 50 -c 10 $URL
(ab is a benchmark tool provided with Apache2)
Ran some iterations until results became stable enough.
Results:
PHP4-CGI: 4,2#/sec at avg. 240ms (100%)
PHP4-FastCGI: 5,6#/sec at avg. 177ms (133%)
PHP5-FastCGI: 5,0#/sec at avg. 200ms (119%)
PHP5-FastCGI+APC: 14,8#/sec at avg. 68ms (352%)
Interpretation:
PHP5 is indeed slower than PHP4.
FastCGI seems faster than CGI, even for complex PHP applications.
APC can speed up UNB to 3x.
Maybe some background on the PHP installation:
PHP4-CGI is PHP 4.4.1 invoked through suPHP to get the script running in the correct user context. Configuration changes: register_globals, register_argc_argv. No further extensions.
PHP4-FastCGI is the same PHP, only invoked through Apache's suEXEC and mod_fastcgi.
PHP5-FastCGI is PHP 5.1.1 invoked through the same mod_fastcgi. Configuration changes: register_argc_argv. No further extensions.
PHP5-FastCGI+APC is the same PHP5-FastCGI, only with the APC extension loaded and active, with the APC optimiser disabled. This caches the compiled intermediate code inside PHP and thus saves parsing the large codebase on every request. This is the most impressive effect of the whole test.
♪ ...nanananah, all in all we’re just brilliant thieves, nanananah... ♪♬
This post was edited on 2006-01-04, 23:56 by