How to optimize the php process memory usage?

10

I am running a wordpress site and each PHP process usage about 200mb to 250mb resident size memory. With 16GB of ram, the server can only handle about 70 processes. By increasing the virtual memory to 16GB it can handle 140. After that the load keeps rising. If there are 200 connections in 10 minutes, the server load reaches 20 on a 3Ghz quad-core xeon processor!

I have tried deactivating all the plugins, but this only reduces the PHP memory usage of each process by less than 10%. suPHP tells me which user is using so much memory, but not what part of the wordpress code.

Any suggestion on how to reduce the memory usage? Or is my only option to upgrade to 32GB of ram?

PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
10585 nobody    16   0 2266m 237m 199m S 21.3  1.5   1:09.17 /usr/bin/php
10597 nobody    16   0 2257m 255m 226m S 15.3  1.6   0:17.56 /usr/bin/php

Biggest outputs from pmap -d

000000000e8b8000   27580 rw--- 000000000e8b8000 000:00000   [ anon ]
00002b3772850000 2097152 rw-s- 0000000000000000 000:00009   [ shmid=0x2d1b803a ]
00002b37f2a62000   55108 r---- 0000000000000000 0fd:00000 locale-archive
mapped: 2320852K    writeable/private: 30012K    shared: 2097152K

ipcs output

------ Semaphore Arrays --------

key        semid      owner      perms      nsems
0x000000a7 0          root      600        1
0x00000000 162529281  nobody    600        1
0x00000000 162562050  nobody    600        1
0x00000000 162594819  nobody    600        1
0x00000000 162627588  nobody    600        1
------ Message Queues --------

key        msqid      owner      perms      used-bytes   messages`
php
linux
wordpress
centos
asked on Stack Overflow Apr 27, 2012 by Lisa • edited Mar 16, 2014 by hakre

2 Answers

7

I'll summarize what Lisa did to find the problem:

  • Check the memory layout of a single PHP process with pmap -d <pid>. The output showed that there's a huge amount of shared memory used by the process:
00002b3772850000 2097152 rw-s- 0000000000000000 000:00009   [ shmid=0x2d1b803a ]
  • Examine the shared memory regions with ipcs -m. It showed that there are a lot of shared memory regions created by user nobody (the web server), here are just a few of them:
0x00000000 117964807 nobody 600 2147483648 1 dest 
0x00000000 117997576 nobody 600 2147483648 1 dest 
0x00000000 118030345 nobody 600 2147483648 1 dest
0x00000000 118063114 nobody 600 2147483648 1 dest
  • Disable eAccelerator in php.ini and remove the created shared memory regions:

for i in `ipcs -m | cut -d' ' -f2 | grep '^[0-9]'`; do ipcrm -m $i; done

answered on Stack Overflow Apr 27, 2012 by strkol
3

Rasmus Lerdorf did a conference about PHP performance at Confoo in 2010 and he used a Wordpress blog as an example, this should give you great tools to answer your question:

http://talks.php.net/show/confoo10/1

To sum up:

  • Run a phpinfo() and disable PHP extensions that you don't use. They can take a lot of memory (imagick, curl, ...)
  • Generate a graph of your includes using the inclued.so extension. You might load useless functions in your wordpress setup.
  • Run benchmarks with siege. Sometimes, tiny optimisations have great impact on performance, so make sure you have metrics, to help you make your decisions.
  • Use callgrind to show where you're loosing performance. In one of my project I was using md5() to hash my SQL queries and cache them. The md5() calls where using 20% of the CPU time.

I would definitely start by disabling PHP extensions if possible.

answered on Stack Overflow Apr 27, 2012 by Tchoupi

User contributions licensed under CC BY-SA 3.0