2

Has anyone else experience this issue with HTTP/2 on IIS10. I have a VM set up on Azure with a Website on IIS10. I've set up SSL etc. and tested the site on HTTP/2, turned off SSL and tested it on same browser with HTTP/1.1. The result was unexpected, HTTP/1.1 outperformed HTTP/2 BIG TIME!

Tests were done from same machines, exact same webpage (contains lots of resources to load) and the only thing that changed was the protocol.

HTTP/2 results: HTTP/2 Loading times

HTTP/1.1 results: enter image description here

As you can see HTTP/1.1 outperforms HTTP/2 and to me it looks like the server is taking an extraordinary amount of time to respond to the requests (green - Time to First Byte).

Any ideas?

  • One hypothesis is that IIS10 and Windows Server 2016 are not finished yet and there are problems with HTTP/2 performance but I could not find anyone else complaining about it – Daniel Wardin Jun 23 '15 at 10:04
  • I am using unencrypted HTTP/1.1 against encrypted HTTP/2. Do you think that makes a difference, should unencrypted HTTP/1.1 outperform encrypted HTTP/2? – Daniel Wardin Jun 23 '15 at 14:50
2

The difference is in how HTTP/1 and HTTP/2 set up sockets for networking.

In HTTP/1, the browser makes a request for a single asset and then waits for that asset to download before making the next request. This is very slow, so to get around that, browsers started multiplexing their requests and making 6-8 requests at a time, splitting the bandwidth between those 6-8 requests.

In HTTP/2, the browser opens a single socket, makes all of the requests and then waits for all of the assets to be streamed at the highest available bandwidth.

The reason that you're supposed to see a high TTFB is because in HTTP/1, each request is a separate socket, but in HTTP/2, all requests download through the single stream. You may have made the requests, but you have to wait for the server to respond with all of the data that you've requested.

In theory, HTTP/2 is supposed to be faster because it's more efficient - you don't have the overhead of opening a new socket for each request. You're streaming all of the data down a single data pipe and you can prioritize requests based on how the page should load.

In practice, this depends upon the browser making prioritized requests properly and the network allowing full bandwidth for the single socket stream and not limiting requests artificially.

Additionally, since you've stated that you're doing unencrypted HTTP/1 and encrypted HTTP/2, encryption will cause additional overhead and performance loss. You might try doing encrypted HTTP/1 for a better comparison.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.