im currently testing centminmod for a new server build that will ultimately host a new xenforo community, a fairly busy vB community (60k users, 16.5+M posts), and a few other minor odds and ends.
performance with centminmod isnt too bad. ive been able to hit :
Requests per second: 87663.63 [#/sec] (mean) with static content and Requests per second: 5686.19 [#/sec] for php content.
The only caching involved is APC/memcache as installed by centminmod. No varnish, no cloudflare, etc. The server is a decent box, 4x64g SSD in Raid5, 24gig DDR3 memory, Dual X5675 Hexcore CPUs @ 3.06GHz - 24 cores with HT enabled. I was hoping to get over 100k req/sec with static but couldnt break thru 90k. Should I expect a heck of a lot more performance out of this box? A default install of Xenforo nets around 600 req/sec at the homepage of the install. Was expecting much higher with APC in the mix.
That brings me to my centmindmod concern, why is a newer version of php bundled, why is suhosin still being used(it hasnt been updated in FOREVER), and why is the new Opcode cache contributed by zend not being used versus the deprecated APC? Centminmod has newer version of MariaDB but still a much older ver of PHP.
Suhosin isn't installed by default at all, so it shouldn't have been compiled ?? Or you mean the menu option for Suhosin to be installable by choice if user wants to ?
Reason for PHP 5.3 is actually listed on the web site
http://centminmod.com/phpfpm.html. There's no guarantee that XF add ons are all fully PHP 5.4/5.5 supported. Not everyone just runs XF standalone, for alot of folks, I am sure other PHP scripts are used on same server. So PHP compatibility is left to each user to decide and evaluate against.
Like Andy, I have Centmin Mod with PHP 5.4.18 right now and APC Cache. From my own experience, I get better performance with APC 3.1.13 + igbinary default in Centmin Mod vs PHP 5.4/5.5 Zend OPCache.
If you run centmin.sh and menu option 5 to upgrade PHP and enter PHP 5.5.2, you will be asked if you want to use Zend OPCache or stick with APC 3.1.13 + igbinary. So you can test either method
Although, I have reports of folks having instability with Zend OPCache with some PHP scripts.
As to PHP performance, what benchmarking tools you use ? siege benchmark (which is built in and installed in Centmin Mod by default i.e.
http://centminmod.com/benchmarks.html) or apache bench ? For your requests per second format looks like apache bench used ?
If you are benching against PHP with MySQL backend, then you are partially testing MySQL backend performance as well. So you need to tune for PHP and MySQL as well. What was the exact apache bench command and testing parameters used ?
As to performance, you have 24 cpu threads. So you might want to tune nginx.conf and php-fpm conf via respective short cut commands, nginxconf and fpmconf.
Default Centmin Mod only uses 1 worker_processes as the original intention was for low memory servers <512MB VPS usage which usually only come with 1-2 cpu threads. And default php-fpm conf uses ondemand and low number of pm.max.children and pm.max_requests value. You can through trial and error find more optimal values suited to your server hardware i.e balancing number of worker_processes and pm.children amongst your 24 cpu threads.
Also requests per second value is also a function of the test parameters used for apache bench or siege benchmark.