Frontend Performance

After happening on Ilya Grigorik’s blog where he writes about his work with the Make The Web Fast Team at Google, I found a great talk on web performance. Here are some notes I’ve taken after watching the talk.

Key Takeaways

• Latency is more important than bandwidth in determining page speed.

• The average website is 1MB+ and requires 80+ different resources, yet the average RTT between client and server in the US can be up to 250ms (400ms on a 3G network). It seems like an impossible task to make one’s site load within 250ms (which is the magic number at which you feel the response “feels fast”)

• Currently Chrome uses 6 TCP connections per host, and stops at 6 because any higher and performance would start to go down due to TCP’s “slow start” feature where the first request starts with a small congestion window and grows with each subsequent request.

• Mobile will surpass desktop very soon, but latency is even worse on mobile.

• GAQ can track page load time (and uses in-browser implementation of the W3C Navigation Timing spec), but defaults to sampling 1% of connections, so use the following to increase that: