WordPress W3 Total Cache and Alternative PHP Cache (APC)

APC definitely needs tweaking. For a site the size of the one you describe, you will likely need to raise apc.shm_size to prevent cache fragmentation and errors.

I have a blog post about tweaking APC for WordPress and W3 Total Cache, you might like to read it. My post is mostly concerned with getting APC to work with multiple virtual servers running fcgid, but there is quite a bit on APC performance tuning as well.

Most likely APC is conflicting with something. Do you have Zend Optimizer installed? If so that will cause your server to throw a 500 error when you try to load a page.

@c3mdigital: APC is still running, it’s just not being used by the W3TC plugin at the moment and no, Zend Optimizer is not installed. Although it has ionCube running.

More details I would like to add about the site:

The site receives around 600,000 unique visitors per month and around 2 million page views per month.

The details pertaining to the server are as follows:

Processor: Core2Duo E7400 2×2.93GHz 3MB Cache

RAM: 8 GB DDR2

HDD: 500 GB

I am trying to assess if whether the problem in hand is more server related than anything else and if so, what kind of server or server set up would be required at the very minimum?

Thanks for all the replies so far!

That’s a load of traffic, but 8GB ram should, in theory, handle it. However, this requires optimizing Apache and MySQL. The server must provide a sustainable number of user sessions without going into swap memory or disk storage of common queries, tables, etc. It must also quickly provide new connections and recycle idle connections.

If the server “crashes” and cannot allocate memory, Apache and MySQL are likely not optimized.

Here is good advice on tweaking Apache performance to provide a sane number of connections.

For reference, here is an Apache configuration that works well on a 6GB RAM VPS:

httpd.conf settings

TimeOut 300
KeepAlive on
MaxKeepAliveRequests 500
KeepAliveTimeout 10
<IfModule prefork.c>
StartServers           25
MinSpareServers        25
MaxSpareServers       100
ServerLimit           384
MaxClients            256
MaxRequestsPerChild 10000
</IfModule>

MySQL cache is somewhat difficult to tweak. A properly configured MySQL cache can provide a foundation to help W3TC and APC function. Use the MySQLTuner perl script to test the effectiveness of MySQL setup & provide guidance.

For reference, here are MySQL settings that work well on a 6GB RAM VPS server:

my.cnf settings

[mysqld]
set-variable = net_buffer_length=8M
set-variable = max_allowed_packet=2M
set-variable = myisam_sort_buffer_size=72M
set-variable = sort_buffer=8M
set-variable = key_buffer=72M
set-variable = table_cache=3333
set-variable = query_cache_size=256M
set-variable = thread_cache_size=8
set-variable = innodb_buffer_pool_size=28M
set-variable = join_buffer_size=12M
set-variable = tmp_table_size=32M
set-variable = max_heap_table_size=28M
set-variable = max_connections=200
set-variable = wait_timeout=10
set-variable = max_connect_errors=100

ALSO: use stateful iptables firewall & intrusion detection software to lessen wasted resources consumed by bad bots, DOS attacks, or over allocation of connections to crawlers/spiders/bots. Configserver Firewall or OSSEC works well.

The opcode cache you choose is not the issue. I think the keep alive setting being on without using a reverse proxy is part of the problem. eAccelerator doesn’t suport user caching of files so it cannot work with your object cache. I would be curious to learn what the errors are in the error log when things crash.