How to make cache work for you

Sergey Chernyshev

New York, NY

December 10^th, 2009

December '09 NY Web Performance Meetup

What Is Cache?

Fastest way to do something is not to do that at all!
Save the result and reuse it
What to save and when?
For how long?

Cache Layers

Back-end
- Code compilation (opcode cache, JIT)
- Database (SQL Query cache)
- Application (memcached and alike)
- Edge (reverse cache proxies)
Client
- Browser cache
- HTML 5 Application cache and local storage

Back-end

SQL Query Cache

Databases can be smart enough not to re-run queries
Costs resources for each query
Good in theory - test if it works for your data

Fun facts for MySQL query cache

query cache is not enabled by default (query_cache_size = 0)
can use on demand (query_cache_type = 2 and SELECT SQL_CACHE)
not used for prepared statements for MySQL <5.1.17
cleared on update or insert

App Data Cache

Generic cache in your app - tons of different tools

shared memory (ASP.NET, Java)
file system
key-value store
database

App Data Cache: File System

Good for cacheing data fetched from network

"Exclude yourself" pattern

Generate a file for your web server to serve

RewriteCond %{REQUEST_FILENAME} !-f
RewriteRule ^(.+)$ generate.php?filename=$1 [L,QSA]

Similarly, Smarty generates PHP code out of templates and include() them

App Data Cache: Key-Value Store

Huge associative array persisted across requests

$memcache = new Memcache;
$memcache->connect('localhost', 11211)
	or die ("Can't connect");

$val = $memcache->get('hardtocompute');

if ($val === FALSE) {
	$val = computeThisHardThing();
	$memcache->set($val);
}

echo "This one is hard to get, but we cache it: $val";

memcached, APC, Redis, Tokyo Cabinet / Tyrant ... even Amazon S3

Key-value interface can be abstracted (Zend_Cache)

App Data Cache: Conclusion

Requires application changes
You have most control here
Do what makes sense for your app
Handle it in your data model layer

Reverse Cache Proxies

Remove load from your application servers
Easy and powerful cache layer outside of your app
Can be also load balancers
You can use PURGE /image.gif HTTP/1.1
Not the same as CDN
Squid, Varnish, vendors like Akamai EdgeSuite

Client

Rule 1

Rule 1: Make fewer HTTP Requests (t-shirt)
Amazon: 160 requests
CNN: 147 requests
Google: 12 requests

Use Browser Cache

Don't have browser caching issue, use browser caching functionality
All static content can be cached, even HTML
Avoid unnecessary 304s - use far-future expires (last-modified is not enough)
- But, you need to keep URLs unique (CMS changes)
- MediaWiki patch at work on TechPresentations.org: 2 requests on repeat view
Makes sense if pages / visit > 1
- Amazon: 160 requests -> 41 requests
- CNN: 147 requests -> 20 requests
- Google: 12 requests -> 4 requests

HTML5 Application Cache

Caches versions of your web app in the browser
Separate from browser cache
Events can be scripted with JavaScript
Backwards compatible
Works with iPhone / iPod web apps now!

HTML5 Application Cache

On HTML page

<!DOCTYPE HTML>
<html manifest="my.cache.manifest">
...

Manifest file (Content-type: text/cache-manifest)

CACHE MANIFEST
index.html
help.html
style/default.css
images/logo.png
images/backgound.png

HTML5 Local Storage

Same as App data cache, but in browser now
Comes in two flavors:
- sessionStorage
- localStorage
"No network" cache
localStorage with app cache, makes "thick client"
Works with iPhone / iPod web apps now!

HTML5 Local Storage

localStorage.setItem(key, value);
localStorage.getItem(key);
localStorage.removeItem(key);

localStorage['key'] = value;
localStorage.key = value;

localStorage.length;
localStorage.clear();

Conclusion

There are many ways to not do things over and over
When someone is saying "cache", ask what they mean
Cache saves cash!

More info

Varnish - A State of the Art High-Performance Reverse Proxy by Artur Bergman of Wikia (and LiveJournal prior to that)
HTML5 Part Three: Session and local storage, Web databases and application cache (video) by Marcus Lofthouse @ Web Standards Meetup, November 19, 2009
TechPresentations.org/Performance

About these slides

These slides use an HTML presentation tool Slidy.

They are valid XHTML document with rich metadata embedded using RDFa.

Photos used

"Server Rack" photo by Jamison Judd

"Web Browsers 2009 (Map)" photo by tedion

Licensing & Attribution

How to make cache work for you by Sergey Chernyshev
is licensed under a Creative Commons Attribution 3.0 United States License.
Permissions beyond the scope of this license available here.

How to make cache work for you

Sergey Chernyshev

New York, NY

December 10th, 2009

December '09 NY Web Performance Meetup

What Is Cache?

Cache Layers

Back-end

SQL Query Cache

Fun facts for MySQL query cache

App Data Cache

App Data Cache: File System

App Data Cache: Key-Value Store

App Data Cache: Conclusion

Reverse Cache Proxies

Client

Rule 1

Use Browser Cache

HTML5 Application Cache

HTML5 Application Cache

HTML5 Local Storage

HTML5 Local Storage

Conclusion

More info

About these slides

Photos used

Licensing & Attribution

December 10^th, 2009