===== Drupal HTTP Cache =====

[[http://drupal.org/|Drupal]] doesn't support HTTP caching out of the box like [[http://plone.org/documentation/tutorial/optimizing-plone| Plone]], but it can be extended for this, using a small patch and some [[http://apache.org/| Apache2.2]] configuration.

==== Apache 2.2 Configuration ====
First, the apache configuration file ''httpd.conf'' needs to be configured to allow HTTP caching via [[http://httpd.apache.org/docs/2.2/mod/mod_cache.html|mod_cache]] and [[http://httpd.apache.org/docs/2.2/mod/mod_disk_cache.html|mod_disk_cache]].

    # configure disk caching directory
    CacheRoot               /var/cache/apache2/mod_disk_cache/WEBSITENAME
    # enable caching for topmost URLs
    CacheEnable             disk /
    # disable caching for some sub URLs
    CacheDisable            /pipermail

==== Drupal-5.3 Modifications ====
Second, the attached diff {{drupal/httpcache-5.3.diff}} needs to be applied, it changes a couple things:
  - Allow specification of ``anon_http_cache_seconds`` in the config section of ``drupal-5.3/sites/default/settings.php``. A value >0 causes anonymously accessed pages to be HTTP-cached for this number of seconds.
  - The logic in ``drupal-5.3/includes/bootstrap.inc`` is adapted to avoid always forcing html page expiration and allow HTTP caching for ''anon_http_cache_seconds'' seconds.
  - A rewrite rule is added to ``drupal-5.3/.htaccess`` to allow caching of the Drupal root URL.
  - Also, ``bootstrap.inc`` is adapted to avoid caching if the ``X-Script-Chrome:`` header is specified, this is needed in case a theme is used that changes look in correspondence to this header.

==== Benchmarking ====
To verify the usefulness and quantify the gain of the above caching setup, the [[http://httpd.apache.org/docs/2.2/programs/ab.html | Apache Benchmarking Tool]] can be used, for example like this:
    ab -n 100 http://testbit.eu/book/Rapicorn-Tutorial

The first benchmark is executed with Drupal caching completely turned off (adjustable under drupal-5.3/admin/settings/performance). This means that each page is rerendered completely in PHP:
    # Caching-mode:Disabled Aggregate-and-compress-CSS-files:Disabled
    Time taken for tests:   81.608534 seconds
    Complete requests:      100
    Total transferred:      843900 bytes
    HTML transferred:       780800 bytes
    Requests per second:    1.23 [#/sec] (mean)
    Time per request:       816.085 [ms] (mean)
    Transfer rate:          10.10 [Kbytes/sec] received

Then with normal Drupal caching enabled, cached pages are delivered from a page-cache MySQL-DB table with very little PHP code executed:
    # Caching-mode:Normal Aggregate-and-compress-CSS-files:Enabled
    Time taken for tests:   8.862656 seconds
    Complete requests:      100
    Total transferred:      767100 bytes
    HTML transferred:       706200 bytes
    Requests per second:    11.28 [#/sec] (mean)
    Time per request:       88.627 [ms] (mean)
    Transfer rate:          84.51 [Kbytes/sec] received
Page delivery is roughly 9 times faster now.

Finally, the HTTP cache is enabled e.g. by a value like ''anon_http_cache_seconds = 60'' in ''settings.php'' and by ''CacheEnable'' in the Apache configuration:
    # Caching-mode:Normal Aggregate-and-compress-CSS-files:Enabled CacheEnable-disk-/
    Time taken for tests:   0.264496 seconds
    Complete requests:      100
    Total transferred:      769600 bytes
    HTML transferred:       707900 bytes
    Requests per second:    378.08 [#/sec] (mean)
    Time per request:       2.645 [ms] (mean)
    Transfer rate:          2839.36 [Kbytes/sec] received
With Apache handling the majority of page deliveries now, delivery is sped up by a factor of 300, compared to the cacheless setup (and still 33 times faster than with normal Drupal caching).

==== Side Effects ====
It should be noted that Drupal modules (or PHP code in general) that takes action upon equal anonymous page requests will be called less often with activated HTTP caching. That is because Drupal isn't even activated when cached pages can be served by Apache. In practice, no significant problems are expected from active HTTP caching, but for examples page counters can be too small by an order of magnitude. If accurate page access numbers are needed, they should be generated from statistics extraction from Apache's log files (cached page delivery is still regularly logged by Apache).