-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

<p>
I've been working on Angerwhale some more, and now it's
about <b>17</b> times faster than it was before!
</p>
<p>
I achieved this speed increase by changing the way caching works.
Before, I cached pages based on the md5sums of all the visible
articles on the page (along with a list of all tags and categories,
since those show up on the sidebar of every page).  This was pretty
slow because every article and comment had to be enumerated, sorted by
date (to determine which were recent enough to show up on the page),
and then finally md5'd.  This worked pretty well, because a lot of the
time was spent HTML-izing the data (stash) in the TT template.
However, determining which cache key to use still took a significant
amount of time.
<p>
Now, I'm doing things differently.  In the <tt>auto</tt> method,
before most of the application executes, I determine the "revision
number" of the blog.  The revision number is increased (not
necessarily by one) every time the blog is modified -- an article is
tagged, a category is added, a new post appears, etc.  If I have a
cached version of the requested URL for the current revision, then I
immediately give that to the user, along with any saved headers.  This
happens in about 20 miliseconds on my machine.  No request logic is
executed; if there's a cached copy, it's echoed to the user and the
request ends inside <tt>auto</tt>.
</p>
<p>
If I don't have a cached copy, I save the revision number and URL in
the stash (as the "cache key"), run the request as normal, and then in
the <tt>end</tt> action, I save the body and headers to the cache.
</p>
<p>
That's pretty much all there is to it.  I'm also saving the
last-modified time and an e-tag for every entry so that browsers and
RSS readers can issue conditional GET requests.  This saves bandwidth
in addition to CPU time.  If the RSS reader already has the latest
feed, I return a "304 Not Modified" response and don't even have to
serve the cached feed to the RSS reader.  (This works with browsers,
too.  Try firing up Angerwhale and clicking reload constantly.  On
every request after the first, your browser will get "304 Not
Modified" and load the document from cache instead of the web server.
Fast!)
</p>
<p>
(There is one disadvantage -- I have to maintain a cached copy of
every page for every logged in user because the user's name shows up
in the sidebar when he's logged in.  I plan to fix this by using
javascript to render the username on the client side.  If you don't
have javascript, you'll just have to remember your name.)
</p>
<p>
In case you're wondering, here are the <tt>ab</tt> results for the
main page.<br />
Before:
<pre>
Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:        0    0   0.0      0       0
Processing:   835  844   6.7    847     855
Waiting:      832  841   7.0    844     853
Total:        835  844   6.7    847     855
</pre>
<br />
After:
<pre>
Connection Times (ms)
              min  mean[+/-sd] median   max
Connect:        0    0   0.0      0       0
Processing:    45   48   3.0     48      56
Waiting:       42   45   3.4     46      54
Total:         45   48   3.0     48      56
</pre>
<br />
This is all with the same data, but served with the new caching
algorithm.  Pretty cool, eh!?
</p>

<p>
Anyway, if you're using Angerwhale, I hope you enjoy the speed
increase.  Let me know if you notice any issues!
</p>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)

iQCVAwUBRb3Jd9AZeFPdJeQvAQLu7AQAlByIFzhoU+0ILxwuTCzUdZ40FC1CjKj3
eqLP+4GpWBWh9q7nWn92ihpFW1/t+iZGaK/NPCJzlKUgcV2I3BFcIx0HDNfixOAb
nd36AKKdVWGHgG+cX+oS3dzby4SZEhBid3XflxilKsRvkJDFCowGbon7PS7dC4hg
wn8AXpLSWrQ=
=rg++
-----END PGP SIGNATURE-----