Cloudfront performance testing


We're in the process of rolling out Amazon Cloudfront on our systems as the next step in enhancing performance for our customers and users.

While I was confident that it would make a difference I wanted to get an understanding of exactly how much of a difference it would make to an end user, so I performed to basic A/B testing of page load with and without the CDN enabled.

The testing was performed from our Sydney office using Amazon Cloudfront which has a CDN Point of Presence in Sydney, and the main application was running from our staging environment in Amazon's North California data center.

The results were significantly better than I was expecting and are best illustrated by the picture below:

Chart of performance of CDN vs No CDN

Key findings:

  • Time to load complete was reduced by 35% (~2 seconds)
  • All of the gain was in the time between DomReady and the Onload firing
  • It's definitely worth the effort

The other key question that was asked around the office was is it worth the effort given it's only on first page load and we aggressively cache all the static assets, so for most users it won't matter.

The two counter arguments to this are:

  • First impressions count – so every page load should be fast
  • As this is a SAAS model we do upgrades frequently – which even for existing users will invalidate their cached data

So, we'll definitely be rolling it out over the coming weeks and months to more and more of our systems.


This post was posted in , , , , by on