cURL / Mailing Lists / curl-library / Single Mail


Re: HTTP 2.0 performance is slower than HTTP 1.1 in libcurl

From: Daniel Stenberg <>
Date: Thu, 19 May 2016 12:00:05 +0200 (CEST)

On Thu, 19 May 2016, Ganesh Nikam wrote:

> Looking at all the documentation available about HTTP 2.0, it seems that it
> should perform better than HTTP 1.1 (Here we are measuring performance on
> the basis of time taken to perform X number of request to particular web
> server).

I don't think we can expect that it always will perform better, but in most
cases it certainly should. Assuming HTTPS, the same ciphers, similar network
sitation etc.

But as I just mentioned the other day: we have not yet had the change to work
a lot on optimizing our h2 transfers so I'm sure we're not doing everything
the best possible way yet. Maybe we're even doing plain stupid things.

Getting (reproducable) testing going is a great first step. Rajiullah Shammi
posted a great test setup the other day, basically saying the same thing you
say (on a different test scenario):

Next we need to do careful traces of libcurl runs to see where the time is
spent (wasted) and what to do about it.

> Internal conversion of the response header (from HTTP 2.0 to HTTP 1.1) make
> HTTP 2.0 slower ?

I doubt that particular functionality has any measureable impact, but as I
said: we need to measure and trace to understand where we need to improve.

List admin:
Received on 2016-05-19