07 Ott 11 Apache with Worker MPM (multi threaded), mem_cache and mod_deflate

When you have to publish mainly static contents, like static sites, the most powerful solution is to configure your apache http server to use the MPM Worker, mod_mem_cache e mod_deflate modules.

Why the MPM Worker
It implements a multi process / multi thread server model. The father process spawn processes, and each child process spawn threads. Each thread will handle a client connection.
This implementation can handle a large number of requests with fewer system resources than a standard prefork multi process server model.
Please note that you cannot use the MPM Worker in server environments that are not thread safe. For example, PHP, mod_perl, and other dynamic page processors do not ensure you that the environment it’s completely thread safe, so my advice is to NOT USE the MPM Worker with PHP, mod_perl and the like.
The Worker MPM can consume much less memory because the heap memory is shared among threads, while that’s not true for processes.
For more informations you can read the official page: http://httpd.apache.org/docs/2.2/mod/worker.html

Why the mod_mem_cache module
This module can be configured to cache open file descriptors and objects into the heap storage (memory).
If the same object (html, css, js, etc) it’s requested for the first time by a client, it get saved into the heap memory. The second time it got requested, the object got feeded directly from the memory cache. It can lower down CPU and disk I/O.
For more informations you can read the official page: http://httpd.apache.org/docs/2.2/mod/mod_mem_cache.html

Why the mod_deflate module
It can allows output from your server to be compressed before being sent to the client . The HTTP 1/1 protocol has a header called Accept-Encoding. This way a client can tell the server witch response encoding it can reads.
Any modern browsers today can handle page compression, so why not using it?
With it you can save bandwidth.
For more informations you can read the official page: http://httpd.apache.org/docs/2.2/mod/mod_deflate.html

Ok. Let’s begin to enable that stuff.

First step is to compile apache from source.
If you want to use the packages released by your linux distribution instead of compiling apache by yourself you can do it.
Always choose the latest apache stable version available.

To compile apache 2.2.X with most modules in shared form (*.so) you should run this configure:
$ ./configure –prefix=<YOUR_APACHE_DIR> –with-mpm=worker –with-included-apr –with-expat=builtin –enable-mods-shared=most –enable-ssl –enable-proxy –enable-proxy-connect –enable-proxy-http –enable-proxy-balancer –enable-cache –enable-disk-cache –enable-mem-cache –enable-nonportable-atomics=yes

Then, as usual, run:
$ make
$ make install

You hopefully end up with apache correctly installed with all needed modules in place.
Now configure your httpd.conf adding those lines:

# Compress on the fly HTML pages, TXT and XML files, CSS and JS.
AddOutputFilterByType DEFLATE text/html text/plain text/xml text/css text/x-js application/x-javascript

# Cache open file descriptors
CacheEnable fd /

# Enable memory caching
CacheEnable mem /

# Limit the size of the cache to 24 Megabyte
MCacheSize 25165824

# Minimum size of an object that can be cached: 1 Kbyte
MCacheMinObjectSize 1024

# Maximum size of an object that can be cached: 3 Mbyte
MCacheMaxObjectSize 3145728

# Spawn 10 child processes, spawning 100 threads for each child process.
# So, a pool of 1000 threads is left up and sleeping, ready to serve incoming requests.
# If more requests will come in, apache will spawn new child processes, each one spawning 100 threads,
# enlarging the thread pool until the total number of threads become 2000. In that case, apache begin
# to cleanly drop processes, trying to reach 1000 threads.
# New processes and its threads are spawned in case of a large spike of requests, until 4000 parallel
# client requests are reached, then apache will no longer accept new incoming connections.
# When the load calm down, and requests come back under 4000 parallel connections, apache will continue
# to accept connections. After 1,000,000 requests served by a child, q. 10,000 per thread, the process
# get closed by the father to ensure no memory leak is fired.
<IfModule mpm_worker_module>
ThreadLimit          100
ServerLimit         4000
StartServers          10
MaxClients          4000
MinSpareThreads      1000
MaxSpareThreads      2000
ThreadsPerChild      100
MaxRequestsPerChild   1000000
</IfModule>