FastCGI vs FPM results

[Chris1]

Hello all

I decided to make a comparison between a few of the different PHP handlers provided by Plesk 12.5, I thought you may like to see.

• All tests performed on an account limited by CloudLinux 7.1 LVE defaults and also within CageFS.
• Apache Benchmark was performed on the server the tested account resides on.
• The test website is a stock Wordpress 4.3 install.
• All results are in average requests served per second over 10 tests each

ab -n 50 -c 1 (1 user x 50 requests)
5.6.12 FastCGI - 14.08
5.6.12 FPM served by Apache - 14.19
5.6.12 FPM served by nginx - 20.12

ab -n 100 -c 5 (3 users x 50 requests)
5.6.12 FastCGI - 20.27
5.6.12 FPM served by Apache - 48.73
5.6.12 FPM served by nginx - 56.72

ab -n 100 -c 5 (5 users x 100 requests)
5.6.12 FastCGI - 14.47
5.6.12 FPM served by Apache - 35.67
5.6.12 FPM served by nginx - 82.30

ab -n 100 -c 20 (20 users x 100 requests)
5.6.12 FastCGI - 13.58
5.6.12 FPM served by Apache - 57.60
5.6.12 FPM served by nginx - 82.38

ab -n 100 -c 100 (100 users x 100 requests)
5.6.12 FastCGI - 20.62
5.6.12 FPM served by Apache - 70.50
5.6.12 FPM served by nginx - 73.73

 

[UFHH01]

Hi Chris1,

sorry to say, but it's pretty useless to use an apache benchmark test on a test environment without css, images, content and what ever else contained on the URL. I wonder what makes you think that your test result in a decent information to compare the PHP - handlers? Please use for example "http://www.webpagetest.org/" ( or other usefull tests ) to compare.

 

[Chris1]

Regardless of this, there is still a significant difference in what it does test.

Hi Chris1,

sorry to say, but it's pretty useless to use an apache benchmark test on a test environment without css, images, content and what ever else contained on the URL. I wonder what makes you think that your test result in a decent information to compare the PHP - handlers? Please use for example "http://www.webpagetest.org/" ( or other usefull tests ) to compare.

 

[Dx3webs]

thanks.. very interesting.. + I am sure it would be good to see other tests. I suspect the relative differences will be about the same

 

[OlegN]

Thanks for test results. Want to note that Plesk 12.5 shipped with nginx 1.9.2 which has "reuseport" feature https://www.nginx.com/blog/socket-sharding-nginx-release-1-9-1/ which help to serve even more requests with minimal CPU usage. This nginx feature can be used with minimal tuning on OSes with kernel 3.9+ (Centos 7, Ubuntu 14 and Debian 8)

 

[vibhor202]

Thanks for the test results. However, I guess which handler to use more depends on specific use cases. I did some reading and found the reasoning given quite helpful in predicting where to use which handler.
Here is the guide if someone wishes to verify or go through : fastcgi vs fpm

 

[blueberry]

I just wanted to say that I get the same results to display a wordpress page. FastCGI delivers content faster thant php-fpm. Additionnally, I have noticed that the pageload when using fastcgi was more stable than when using php-fpm.
UFHH01 said it was useless to use an apache benchmark test on a test environment without css, images, content and what ever else contained on the URL.
I don't think so because static files are delivered by Nginx, so i guess it would not impact the benchmark.

 

[weltonw]

UFHH01 said it was useless to use an apache benchmark test on a test environment without css, images, content and what ever else contained on the URL.
I don't think so because static files are delivered by Nginx, so i guess it would not impact the benchmark.

Testing assets, ie via wpt vs running a concurrency benchmark are totally different things and both have their validity.

I just wanted to say that I get the same results to display a wordpress page. FastCGI delivers content faster thant php-fpm. Additionnally, I have noticed that the pageload when using fastcgi was more stable than when using php-fpm.

That shouldn't be the case. FPM is FastCGI pooled, which has the benefit of not needing to spin up as many threads and persistent cache. How are they configured? Ie, if max_request per children is configured to 20, FPM will be "unstable" as you mentioned.