# Optimizing worker\_processes, worker\_connections & PHP\_FCGI\_CHILDREN - Any Good Tutorial? **URL:** \ **Category:** NGINX\ **Created:** [June 17, 2010, 4:22pm UTC](https://www.ruby-forum.com/t/optimizing-worker-processes-worker-connections-php-fcgi-children-any-good-tutorial/190218 "2010-06-17T16:22:23Z")\ **Posts on this page:** 15\ **Page:** 2
**Author:** ![rahul286](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/r/d78d45/32.png) [@rahul286](https://www.ruby-forum.com/u/rahul286)\ **Post date:** [June 17, 2010, 6:28pm UTC](https://www.ruby-forum.com/t/optimizing-worker-processes-worker-connections-php-fcgi-children-any-good-tutorial/190218/21 "2010-06-17T18:28:28Z")
On 15.06.2010 22:27, Rahul B. wrote: > > worker\_process = num of CPU. nothing more > > I will make this a law for myself. 🙂 I recommend have few more workers than number of cores. On one of my overloaded dual core server I changed 2 worksers to 6, this reduced a lot IOwait and fixed latency problem when using site which is on this server. Play a bit with it to found best for you number of workers. – Piotr. ---
**Author:** ![rahul286](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/r/d78d45/32.png) [@rahul286](https://www.ruby-forum.com/u/rahul286)\ **Post date:** [June 17, 2010, 6:28pm UTC](https://www.ruby-forum.com/t/optimizing-worker-processes-worker-connections-php-fcgi-children-any-good-tutorial/190218/22 "2010-06-17T18:28:28Z")
> interesting. Does nginx handle this fcgi response ? You can open a bug > on [bugs.php.net](http://bugs.php.net) to request this feature and, if you have so, propose a > corresponding patch. It has been requested couple of times: > **[PHP :: Bug #39809 :: FastCGI Requests silently dropped](https://bugs.php.net/bug.php?id=39809)** > **[PHP :: Bug #43610 :: fastcgi socket dies on high concurrency](https://bugs.php.net/bug.php?id=43610)** .. then the response (at that moment) that the php master process can’t do that. I wonder now with the use of libevent in the manager if thats still the case. rr ---
**Author:** ![rahul286](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/r/d78d45/32.png) [@rahul286](https://www.ruby-forum.com/u/rahul286)\ **Post date:** [June 17, 2010, 8:36pm UTC](https://www.ruby-forum.com/t/optimizing-worker-processes-worker-connections-php-fcgi-children-any-good-tutorial/190218/23 "2010-06-17T20:36:06Z")
On 15.06.2010 22:48, Rahul B. wrote: > @jabberuser > > > I recommend have few more workers than number of cores. On one of my > > overloaded dual core server I changed 2 worksers to 6, this reduced a > > lot IOwait and fixed latency problem when using site which is on this > > server. Play a bit with it to found best for you number of workers. > > Thanks for suggestions. > I may go you way just in case fastcgi\_cache in nginx improves my > performance. > In that case I will reduce value of PHP\_FCGI\_CHILDREN to offset it. I have there 6 nginx workers and 10 php childrens, restarted after 100 request each PHP\_FCGI\_CHILDREN=10 PHP\_FCGI\_MAX\_REQUESTS=100 Php use ~22M of ram 16.5 MiB + 5.4 MiB = 21.9 MiB php-cgi (11) And nginx ~5M 2.9 MiB + 1.4 MiB = 4.4 MiB nginx (7) (base on ps\_mem.py script). nginx workers is one thing and php another. You host only php or also static files? – Piotr. ---
**Author:** ![rahul286](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/r/d78d45/32.png) [@rahul286](https://www.ruby-forum.com/u/rahul286)\ **Post date:** [June 17, 2010, 8:36pm UTC](https://www.ruby-forum.com/t/optimizing-worker-processes-worker-connections-php-fcgi-children-any-good-tutorial/190218/24 "2010-06-17T20:36:06Z")
@jerome > no i mean memory used by the kernel to cache local files in order not > to access file on hardrive. see output of free, the column “cached” I already have that Cache it seems. Following is my output of Free command. root@rtblogs:~# free total used free shared buffers cached Mem: 1451740 483524 968216 0 61192 283196 -/+ buffers/cache: 139136 1312604 Swap: 262136 0 262136 I must thanks linode for this. But this always reduced my free memory by huge amount. I guess such situations force us to spend more time in deciding on what to cache and how much to cache. We have too many levels where we can cache and good thing about them is that they all happily live together. @jabberuser > I recommend have few more workers than number of cores. On one of my > overloaded dual core server I changed 2 worksers to 6, this reduced a > lot IOwait and fixed latency problem when using site which is on this > server. Play a bit with it to found best for you number of workers. Thanks for suggestions. I may go you way just in case fastcgi\_cache in nginx improves my performance. In that case I will reduce value of PHP\_FCGI\_CHILDREN to offset it. -Rahul ---
**Author:** ![rahul286](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/r/d78d45/32.png) [@rahul286](https://www.ruby-forum.com/u/rahul286)\ **Post date:** [June 17, 2010, 8:36pm UTC](https://www.ruby-forum.com/t/optimizing-worker-processes-worker-connections-php-fcgi-children-any-good-tutorial/190218/25 "2010-06-17T20:36:41Z")
## Piotr K. Wrote: > Php use ~22M of ram > 16.5 MiB + 5.4 MiB = 21.9 MiB php-cgi (11) > And nginx ~5M > 2.9 MiB + 1.4 MiB = 4.4 MiB nginx (7) > (base on ps\_mem.py script). I ran ps\_mem.py [[http://www.pixelbeat.org/scripts/ps\_mem.py](http://www.pixelbeat.org/scripts/ps_mem.py)] And found… 5.5 MiB + 992.5 KiB = 6.4 MiB nginx (9) 49.1 MiB + 60.7 MiB = 109.7 MiB php-cgi (6) I guess for PHP, shared 60.7MB contains APC cache of 60MB. And also total memory used by php-cgi is not 6_109.7 but 6_49.1 + 60 Also my php-cgi is double in size compared to your php-cgi Should I think of removing php extensions in this case! Posted at Nginx Forum: [http://forum.nginx.org/read.php?2,98498,98666#msg-98666](http://forum.nginx.org/read.php?2,98498,98666#msg-98666) ---
**Author:** ![rahul286](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/r/d78d45/32.png) [@rahul286](https://www.ruby-forum.com/u/rahul286)\ **Post date:** [June 17, 2010, 8:36pm UTC](https://www.ruby-forum.com/t/optimizing-worker-processes-worker-connections-php-fcgi-children-any-good-tutorial/190218/26 "2010-06-17T20:36:42Z")
On 15.06.2010 23:28, rahul286 wrote: > 5.5 MiB + 992.5 KiB = 6.4 MiB nginx (9) > 49.1 MiB + 60.7 MiB = 109.7 MiB php-cgi (6) > > I guess for PHP, shared 60.7MB contains APC cache of 60MB. > And also total memory used by php-cgi is not 6_109.7 but 6_49.1 + 60 > > Also my php-cgi is double in size compared to your php-cgi > Should I think of removing php extensions in this case! I don’t know why my php use less memory than yours, I am not a php guy but my server is powered by Funtoo Linux (Gentoo fork) and php is compiled _only_ with what I need, nothing more. – Piotr. ---
**Author:** ![rahul286](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/r/d78d45/32.png) [@rahul286](https://www.ruby-forum.com/u/rahul286)\ **Post date:** [June 17, 2010, 8:36pm UTC](https://www.ruby-forum.com/t/optimizing-worker-processes-worker-connections-php-fcgi-children-any-good-tutorial/190218/27 "2010-06-17T20:36:06Z")
## Piotr K. Wrote: > nginx workers is one thing and php another. You > host only php or also static files? I agree with you completely. I am running a wordpress(es) only as PHP application. All wordpress static content like images, js/css files are served via CDN. So as far as typical static content is concerned nginx have no considerable load to worry about. Then another “static” content comes from output of PHP cached as static html files. During my last test, after “caching” was done, I stopped fastcgi-php & mysql server and as expected request still served by server. Of course back-end was completely down in that case. It is backend and some portion of front-end which amounts to PHP load. And I am putting efforts to make backend faster compared to its current load time. Posted at Nginx Forum: [http://forum.nginx.org/read.php?2,98498,98661#msg-98661](http://forum.nginx.org/read.php?2,98498,98661#msg-98661) ---
**Author:** ![rahul286](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/r/d78d45/32.png) [@rahul286](https://www.ruby-forum.com/u/rahul286)\ **Post date:** [June 17, 2010, 8:36pm UTC](https://www.ruby-forum.com/t/optimizing-worker-processes-worker-connections-php-fcgi-children-any-good-tutorial/190218/28 "2010-06-17T20:36:42Z")
@jerome > I don’t know if it’s possible yet, I have to check it out and think > about it. Since FPM has a different design than the php-f?cgi, it’s > time to make a new request on [bugs.php.net](http://bugs.php.net) for FPM (not for php-cgi) I am using PHP version 5.2.10-2ubuntu6.4 Which is better in this case? php-FPM or php-fcgi? Thanks, Rahul ---
**Author:** ![rahul286](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/r/d78d45/32.png) [@rahul286](https://www.ruby-forum.com/u/rahul286)\ **Post date:** [June 17, 2010, 8:36pm UTC](https://www.ruby-forum.com/t/optimizing-worker-processes-worker-connections-php-fcgi-children-any-good-tutorial/190218/29 "2010-06-17T20:36:42Z")
2010/6/15 Reinis R. [removed\_email\_address@domain.invalid](mailto:removed_email_address@domain.invalid): > that. > > I wonder now with the use of libevent in the manager if thats still the > case. I don’t know if it’s possible yet, I have to check it out and think about it. Since FPM has a different design than the php-f?cgi, it’s time to make a new request on [bugs.php.net](http://bugs.php.net) for FPM (not for php-cgi) ---
**Author:** ![rahul286](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/r/d78d45/32.png) [@rahul286](https://www.ruby-forum.com/u/rahul286)\ **Post date:** [June 18, 2010, 4:39am UTC](https://www.ruby-forum.com/t/optimizing-worker-processes-worker-connections-php-fcgi-children-any-good-tutorial/190218/30 "2010-06-18T04:39:02Z")
## Reinis R. Wrote: > there is no need to recompile script each time a > new child is spawned). Not > to mention other benefits like persistant > connections (to memcache / db / > etc) which aren’t killed each time a request > finishes but reused on next. Hi All New to the forum and new to nginx. Found out about it via the vBulletin forum and am currently setting up a small server to run my vB powered site with 100% nginx & php-fpm. I would love to know more about this caching mechanism, never heard of it before. Are there any concrete examples of its use, sounds perfect for me running on a 768Mb VPS. The current setup I have runs nginx with 2 workers and 2048 connections per worker. I have 20 php-fpm children running with Xcache which serves the site OK until the connections increase. I hit the site with a Loadimpact run and around 120 connection it started spewing out 404’s. The only way to stop it was to restart php-fpm. So the facility above sounds perfect for my limited memory setting… Any help suggestions or pointers would be greatly appreciated. Just want to dump apache as quick as I can! 🙂 Cheer Bruce Posted at Nginx Forum: [http://forum.nginx.org/read.php?2,98498,99692#msg-99692](http://forum.nginx.org/read.php?2,98498,99692#msg-99692) ---
**Author:** ![rahul286](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/r/d78d45/32.png) [@rahul286](https://www.ruby-forum.com/u/rahul286)\ **Post date:** [June 27, 2010, 3:46am UTC](https://www.ruby-forum.com/t/optimizing-worker-processes-worker-connections-php-fcgi-children-any-good-tutorial/190218/31 "2010-06-27T03:46:18Z")
## rahul286 Wrote: > cache levels after that… > -Rahu > Yes! wondering if they all work together. hoping that it will not > sublime it’s ops. Posted at Nginx Forum: [http://forum.nginx.org/read.php?2,98498,102818#msg-102818](http://forum.nginx.org/read.php?2,98498,102818#msg-102818) ---
**Author:** ![rahul286](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/r/d78d45/32.png) [@rahul286](https://www.ruby-forum.com/u/rahul286)\ **Post date:** [June 27, 2010, 1:30am UTC](https://www.ruby-forum.com/t/optimizing-worker-processes-worker-connections-php-fcgi-children-any-good-tutorial/190218/32 "2010-06-27T01:30:31Z")
Thanks All. Moved to PHP-FPM and also removed (purged) unwanted php-extensions. I am using php 5.3.2 and I read somewhere that pm = dynamic feature should be used in PHP-FPM only if PHP \> 5.3.3. Should I wait , or go ahead with Reinis R. settings… > pm = dynamic > pm.max\_children = 70 > pm.start\_servers = 20 > pm.min\_spare\_servers = 5 > pm.max\_spare\_servers = 20 > pm.max\_requests = 1000 I am also thinking of using nginx fastcgi\_cache But started to feel I will be having too many cache levels after that… nginx fastcgi\_cache. apc, memcache and wordpress caching plugins. Will they work all together smoothly or will become counter-productive? Any suggestions? Thanks again, -Rahul Posted at Nginx Forum: [http://forum.nginx.org/read.php?2,98498,102792#msg-102792](http://forum.nginx.org/read.php?2,98498,102792#msg-102792) ---
**Author:** ![rahul286](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/r/d78d45/32.png) [@rahul286](https://www.ruby-forum.com/u/rahul286)\ **Post date:** [September 9, 2011, 2:34pm UTC](https://www.ruby-forum.com/t/optimizing-worker-processes-worker-connections-php-fcgi-children-any-good-tutorial/190218/33 "2011-09-09T14:34:41Z")
Hi, what would you guys recommend to set for a website loaded heavily with PHP and some Ajax features and around 2000 users online at the same time ? Currently I have settings like: PHP\_FCGI\_CHILDREN=10 PHP\_FCGI\_MAX\_REQUESTS=1000 ``` worker_connections 20392; multi_accept off; use epoll; client_body_timeout 200; client_header_timeout 100;. client_max_body_size 40M; client_body_buffer_size 1280k; send_timeout 100; limit_zone limit_per_ip $binary_remote_addr 180m; limit_conn limit_per_ip 150; limit_rate 2200k; keepalive_timeout 15; ``` Using nginx/0.7.67 and php 5.2.17. I cannot migrate to php 5.3.3, so I cannot use native php-fpm. Should I try to tweak fast-cgi better or try to get PHP-FPM working ? Would nginx upgrade help with performance ? Website works but with many users it gets veryyyyyyyyy slow with responses. Thanks guys ! ---
**Author:** ![rahul286](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/r/d78d45/32.png) [@rahul286](https://www.ruby-forum.com/u/rahul286)\ **Post date:** [September 9, 2011, 3:38pm UTC](https://www.ruby-forum.com/t/optimizing-worker-processes-worker-connections-php-fcgi-children-any-good-tutorial/190218/34 "2011-09-09T15:38:49Z")
> I cannot migrate to php 5.3.3, so I cannot use native php-fpm. > Should I try to tweak fast-cgi better or try to get PHP-FPM working ? You can apply FPM patches (it was a quite a long time before it got into core) also to 5.2.x tree and for 5.2.17 ( [http://php-fpm.org/downloads/php-5.2.17-fpm-0.5.14.diff.gz](http://php-fpm.org/downloads/php-5.2.17-fpm-0.5.14.diff.gz) ). The dynamic manager will let you adjust best process numbers and will show the slow scripts. rr ---
**Author:** ![rahul286](https://www.ruby-forum.com/letter_avatar_proxy/v4/letter/r/d78d45/32.png) [@rahul286](https://www.ruby-forum.com/u/rahul286)\ **Post date:** [June 27, 2010, 3:59pm UTC](https://www.ruby-forum.com/t/optimizing-worker-processes-worker-connections-php-fcgi-children-any-good-tutorial/190218/35 "2010-06-27T15:59:04Z")
Hi Eric, > Yes! wondering if they all work together. hoping that it will not > sublime it’s ops. I decided NOT to use memcache. Memcache is good for distributed environment. I am using APC for php opcode cache and user-data as well. In case of w3 total cache plugin, APC can also be used as page-cache instead of hard-disk. Now only confusion, I am in is nginx fastcgi\_cache. In theory, and from what I have read, if nginx fastcgi\_cache serves request, no load should be forwarded to php-fpm or php-fastcgi (in short, php part). I guess with APC also, by using it as page-cache, we are achieving same thing. Not sure though! -Rahul [Previous page](https://www.ruby-forum.com/t/optimizing-worker-processes-worker-connections-php-fcgi-children-any-good-tutorial/190218.md?page=1)