Web Application Performance

Two fundamental limits on performance:

• Amount of data
• Distance (limited by speed of light - fast but not instant)

Other considerations:

• Number of requests - each one involves 1+ round trips (though can re-use connections)

Cache

HTTP headers

• Cache-Control: max-age
• ETag: Hash of file, don't download if unchanged (but beware of how this is calculated, e.g. Apache may use inodes by default which differ on load-balanced servers)
• Expires: Re-download after this
• Last-Modified

Most headers won't help on first visit.

Request headers (conditional GET request):

• If-Modified-Since: Date/time
• If-None-Match: Hash (ETag)

Problems:

• Can't override the browser behaviour - once it's cached with an expiry time

Fingerprinting / hashing

• Gets around caching, effectively forces a new request by using a unique filename based on file contents
• Effectively allows you to set long expiry times but still force updates, e.g. 12 months for site logo

Proxies / intermediates

• Difficult now that everything is HTTPS
• No longer transparent for ISPs etc.
• Can pay a cache provider