Hacker News Reader

Hacker News

Comment by kid0m4n | Hacker News Reader

Full thread

kid0m4n

This is actually pretty impressive!

https://http2.golang.org/gophertiles

View on HN

Noticed that one too. The difference between 1s latency is huge between HTTP/1 and HTTP/2!

Where do you see 1sec latency difference?

EDIT: ok found it and it's striking, the difference is even more than 1 sec (total loading time)

The server doesn't support keep-alive, and it doesn't support pipelining. It seems to be built by design to have the worst possible performance for HTTP/1.1.

Whatever problems pipelining may or may not have, a benchmark like this one loading a bunch of small static images is what it is designed for. I can see how some small-timey Go implementation would not have an actual HTTP/1.1 server, but for Google to come out with HTTP/2 without even having tested against pipelining is just pitiful.

These Google people seem hellbent on not ever having a fair comparison.

Your browser wouldn't even use pipelining if it was supported. The reason HTTP/2 exists is because pipeling ended up not being a solution.

Pipelining works fine. It's on in mobile Safari, and neither Google nor Mozilla found what software it didn't work on (so probably Superfish and other malware). If you use Firefox it's still a great way to speed up normal page loads since HTTP/2 isn't used for unencrypted data.

Pretty sure the demo server supports keep-alive, just as all Go net/http servers do by default: https://github.com/bradfitz/http2/blob/master/h2demo/h2demo....

There's no "Connection: keep-alive" in the response header. If they purposely turned it off that's even worse.

That's because it's the default in HTTP 1.1. The server would need to send "Connection: close" to disable keep-alive.

Apparently "connection: keep-alive" is needed to enable pipelining in the browsers:

http://www.guypo.com/http-pipelining-big-in-mobile/

The comparison there had me literally jumping up and down in my seat.

Reply on news.ycombinator.com