Nginx now most popular HTTPD amongst top 1000 websites

mpkossen

@marcm: exactly! Completely agree with you!

wlanboy

@udk said:
No love for lighty?

Me.

joepie91

@mpkossen said:
OK. But my point was: was it Apache or something else? I mean, like PHP or something else embedded into the apache process. For example: top lists it as apache even when it's a PHP request.

Seeing as I only changed the HTTPd and not the underlying software, and CPU usage magically went down, I can only conclude that it was Apache and not PHP causing the CPU spikes.

@marcm said:
The problem with Apache is that allot of people either don't know how to configure it, are lazy or leave it to its defaults. If you configure Apache right it is going to work pretty damn well.

I always find this a bit of an odd argument. If you need to spend a significant amount of time and have a significant knowledge of a HTTPd to be able to configure it in a way that it actually works properly, then why would you use it if another HTTPd offers this out of the box with no configuration required?

filmgirl

Well I know that at Mashable (where I work as a reporter, not admin, but I'm one of the few that can get excited and need out with our CTO), when we moved from Apache to Nginx, it cut down our overhead and page load like a mofo. And that was back when we were on WordPress on the frontend (now we're on Ruby, Mongo and lots of JavaScript) and needed some ridiculous number of servers and instances just for database.

It doesn't shock me that the bigger sites are switching because if you're serving millions of visitors a day - and often 10,000+ concurrent visitors -- and that's for a dynamic news site, not a web app -- Apache just doesn't work as well.

mpkossen

@joepie91 said:
I always find this a bit of an odd argument. If you need to spend a significant amount of time and have a significant knowledge of a HTTPd to be able to configure it in a way that it actually works properly, then why would you use it if another HTTPd offers this out of the box with no configuration required?

I'm not sure how nginx's defaults are configured, but the Apache ones aren't set for LowEnd by default. So it's only natural you have to tweak it a bit to get better performance. So while I don't disagree with you that Apache doesn't server LowEnd well out of the box, I also don't agree what you need to spend a significant amount of time and have significant knowledge to tweak a number of Apache parameters.

Besides, most people (and most guides) still use the old Apache Prefork server, which is considerably slower than the MPM Worker or the 2.4 Event-driven worker (which matches/is faster as nginx), especially when used with PHP.

Anyway, here's an interesting read about Apache 2.4 and some of the discussion about it's performance: http://apache.slashdot.org/story/12/02/21/1632230/apache-24-takes-direct-aim-at-nginx

marcm

@mpkossen in all fairness I have to say that Nginx is easier to learn than Apache. I find it weird that some people think the opposite.

Swiftlerhost

@marcm said:
mpkossen in all fairness I have to say that Nginx is easier to learn than Apache. I find it weird that some people think the opposite.

The configuration for Nginx is definately easier to learn and use. I would also say performance is better, but I've seen cases where Apache wins.

mpkossen

@marcm said:
mpkossen in all fairness I have to say that Nginx is easier to learn than Apache. I find it weird that some people think the opposite.

I'm all lost in nginx configuration. We have two load balancers here at the office running nginx. Always takes me longer to fix things than with Apache (with which I just know what to do). My point being: it boils down to experience configuration-wise.

vanarp

@mpkossen said:
Besides, most people (and most guides) still use the old Apache Prefork server, which is considerably slower than the MPM Worker or the 2.4 Event-driven worker (which matches/is faster as nginx), especially when used with PHP.

Could you please share any tutorial or tutorial links to best optimize Apache 2.4 for low end boxes?

mpkossen

@vanarp said:
Could you please share any tutorial or tutorial links to best optimize Apache 2.4 for low end boxes?

Haven't got any links for you, sorry. It is on my list for a LEB tutorial, though. I haven't yet had to optimize Apache yet (as in: most of my servers are big enough to handle the load they get just fine). All my servers also still use Apache 2.2, as that's the one that shipped with Ubuntu 12.04 LTS, so I haven't got much hands-on experience with 2.4 yet except some tests.

By the way (edit): the default Apache 2.2 config with the MPM Worker (on Ubuntu) results in a memory usage of at most 6MB reserved per apache process (in my case 4 in total). PHP-FPM, however, is the real memory hog with 3 processes needing 17MB each.

mpkossen

I'd like to get back on this. I've now got two boxes up and running, one with Apache and one with NGINX. Both have:

php5-fpm (tweaked configuration, identical)
postgresql (default configuration, identical)

I've tweaked the default Apache config just a bit (because the default config is not fit for low-end boxes):

<IfModule mpm_worker_module>
StartServers          1
MinSpareThreads      10
MaxSpareThreads      25 
ThreadLimit          25
ThreadsPerChild      10
MaxClients           50
MaxRequestsPerChild   0
</IfModule>

I've compared the memory usage of Apache and NGINX in an idle state:

• Apache: 4 processes, total 9072Kb of residential memory
• NGINX: 5 processes, total 8936Kb of residential memory

This is just a very basic and preliminary result. I've still got to do ab runs against both servers. But I wanted to share this anyhow.

mpkossen

Results from an Apache Bench to a phpinfo() file.

NGINX machine to Apache machine:

mpkossen@postexus2:~$ ab -n 1000 -c 10 http://postexus.example.net/info.php
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/

Benchmarking postexus.example.net (be patient)
Completed 100 requests
Completed 200 requests
Completed 300 requests
Completed 400 requests
Completed 500 requests
Completed 600 requests
Completed 700 requests
Completed 800 requests
Completed 900 requests
Completed 1000 requests
Finished 1000 requests


Server Software:        Apache/2.2.22
Server Hostname:        postexus.example.net
Server Port:            80

Document Path:          /info.php
Document Length:        49595 bytes

Concurrency Level:      10
Time taken for tests:   61.208 seconds
Complete requests:      1000
Failed requests:        0
Write errors:           0
Total transferred:      49787000 bytes
HTML transferred:       49595000 bytes
Requests per second:    16.34 [#/sec] (mean)
Time per request:       612.075 [ms] (mean)
Time per request:       61.208 [ms] (mean, across all concurrent requests)
Transfer rate:          794.35 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:      138  150   6.6    150     165
Processing:   417  459  30.7    455     906
Waiting:      140  157  24.2    155     591
Total:        556  609  35.3    605    1063

Percentage of the requests served within a certain time (ms)
  50%    605
  66%    625
  75%    631
  80%    635
  90%    642
  95%    645
  98%    651
  99%    668
 100%   1063 (longest request)
mpkossen@postexus2:~$

Apache machine to NGINX machine:

mpkossen@postexus:~$ ab -n 1000 -c 10 http://postexus2.example.net/info.php
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/

Benchmarking postexus2.example.net (be patient)
Completed 100 requests
Completed 200 requests
Completed 300 requests
Completed 400 requests
Completed 500 requests
Completed 600 requests
Completed 700 requests
Completed 800 requests
Completed 900 requests
Completed 1000 requests
Finished 1000 requests


Server Software:        nginx/1.1.19
Server Hostname:        postexus2.example.net
Server Port:            80

Document Path:          /info.php
Document Length:        45063 bytes

Concurrency Level:      10
Time taken for tests:   60.874 seconds
Complete requests:      1000
Failed requests:        0
Write errors:           0
Total transferred:      45222000 bytes
HTML transferred:       45063000 bytes
Requests per second:    16.43 [#/sec] (mean)
Time per request:       608.744 [ms] (mean)
Time per request:       60.874 [ms] (mean, across all concurrent requests)
Transfer rate:          725.46 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:      138  150   6.6    150     172
Processing:   418  454  19.7    454     534
Waiting:      139  151   6.9    151     219
Total:        556  605  26.1    604     692

Percentage of the requests served within a certain time (ms)
  50%    604
  66%    623
  75%    629
  80%    631
  90%    639
  95%    643
  98%    646
  99%    647
 100%    692 (longest request)
mpkossen@postexus:~$

Rallias

@mpkossen So you ignore the main strength of nginx?

Try -c 50 or -c 100 if you want to see the real strength of each web server.

InfiniteTech

You can seat Varnish in front Ngnix to drive the performance through the roof. Varnish is great at caching dynamic content. It would serve PHP files as though they were static output.

BluBoy

Would love to see a decent tuning guide for apache and nginx, comparing their strengths and weaknesses as well as a few benchmarks... Aimed at 128/256 Mb RAM nodes would be awesome.

... Anybody in the know with this?

mpkossen

@Rallias said:
mpkossen So you ignore the main strength of nginx?

Try -c 50 or -c 100 if you want to see the real strength of each web server.

I'm an NGINX noob (basically, I have worked with it, but don't know much about it) so I honestly don't know where it's strengths are. I'm going to run the ABs again with higher concurrency, see what that does :-)

@BluBoy: it is on my list for LEB, but these things generally take up a lot of time, so I can't promise it's be finished next week or something like that. More like a month.

mpkossen

Apache:

mpkossen@postexus2:~$ ab -n 1000 -c 50 http://postexus.example.net/info.php
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/

Benchmarking postexus.example.net (be patient)
Completed 100 requests
Completed 200 requests
Completed 300 requests
Completed 400 requests
Completed 500 requests
Completed 600 requests
Completed 700 requests
Completed 800 requests
Completed 900 requests
Completed 1000 requests
Finished 1000 requests


Server Software:        Apache/2.2.22
Server Hostname:        postexus.example.net
Server Port:            80

Document Path:          /info.php
Document Length:        49595 bytes

Concurrency Level:      50
Time taken for tests:   52.008 seconds
Complete requests:      1000
Failed requests:        0
Write errors:           0
Total transferred:      49787000 bytes
HTML transferred:       49595000 bytes
Requests per second:    19.23 [#/sec] (mean)
Time per request:       2600.418 [ms] (mean)
Time per request:       52.008 [ms] (mean, across all concurrent requests)
Transfer rate:          934.85 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:      138  151   7.1    150     209
Processing:   872 2386 186.7   2398    3485
Waiting:      594 2083 183.3   2097    3168
Total:       1011 2536 188.7   2549    3643

Percentage of the requests served within a certain time (ms)
  50%   2549
  66%   2571
  75%   2585
  80%   2592
  90%   2614
  95%   2635
  98%   2712
  99%   3090
 100%   3643 (longest request)
mpkossen@postexus2:~$

NGINX

mpkossen@postexus:~$ ab -n 1000 -c 50 http://postexus2.example.net/info.php
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/

Benchmarking postexus2.example.net (be patient)
Completed 100 requests
Completed 200 requests
Completed 300 requests
Completed 400 requests
Completed 500 requests
Completed 600 requests
Completed 700 requests
Completed 800 requests
Completed 900 requests
Completed 1000 requests
Finished 1000 requests


Server Software:        nginx/1.1.19
Server Hostname:        postexus2.example.net
Server Port:            80

Document Path:          /info.php
Document Length:        45063 bytes

Concurrency Level:      50
Time taken for tests:   12.919 seconds
Complete requests:      1000
Failed requests:        0
Write errors:           0
Total transferred:      45222000 bytes
HTML transferred:       45063000 bytes
Requests per second:    77.41 [#/sec] (mean)
Time per request:       645.928 [ms] (mean)
Time per request:       12.919 [ms] (mean, across all concurrent requests)
Transfer rate:          3418.50 [Kbytes/sec] received

Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:      138  150   6.6    150     166
Processing:   418  485 109.4    457     989
Waiting:      139  181 107.5    153     660
Total:        557  635 110.7    607    1149

Percentage of the requests served within a certain time (ms)
  50%    607
  66%    629
  75%    635
  80%    641
  90%    653
  95%   1012
  98%   1108
  99%   1131
 100%   1149 (longest request)
mpkossen@postexus:~$

I can just say: wow! I'm impressed! Never knew it made that much difference.

I'm curious how this will work with Apache's event worker. I'm trying to set that up now but apt is a bit grumpy this morning.

EDIT: nope, event-worker on 2.2 did nothing. Still the same.

mpkossen

@Zen said:
You have seen the light?

Yes! But I still want to know the difference between NGINX's event model versus Apache's worker model. NGINX doesn't do the keepalive and I want to know what effect that has.

Rallias

@mpkossen said:
Yes! But I still want to know the difference between NGINX's event model versus Apache's worker model. NGINX doesn't do the keepalive and I want to know what effect that has.

Nginx sacrifices a little speed with creating a new connection, Apache sacrifices a lot of speed by being all bloated in the core.