I still see a lot of contradicting benchmark and, apart from some Google Apps, personnally, I have not seen a lot of sites actually really leveraging HTTP2 (including push).

But maybe you did put and leverage HTTP2 on your own website ? At your company ? Did you use push ? Do you use it with CDN ?

Yes, unequivocally. It’s amazing, even without push. The websites that use it are faster, and the development process for making apps or sites that load quickly is much more sane. You don’t have to resort to the kind of weird trickery that pervades HTTP/1 apps.

> Or is it just an improvement for Cloud provider that keep pushing the Kool-Aid ?

I don’t see how that makes any sense at all. Could you explain that?

> But maybe you did put and leverage HTTP2 on your own website ? At your company ? Did you use push ? Do you use it with CDN ?

From my parent comment,

> Speaking as a consumer of the web, as an individual who runs their own website, and at a developer working at a company with a major web presence.

My personal web site uses HTTP/2. It serves a combination of static pages and web apps. No push. HTTP/2 was almost zero effort to set up, and instantly improved performance. With HTTP/2, I’ve changed the way I develop web apps, for the better.

My employer’s website uses every technique under the sun, including push and CDNs.

I've seen a few CDN having a page loading a grid of image in HTTP/1 at page load, and then load the same stuff with HTTP/2 on a button click. It indeed shows you a nice speed up.

Except, when you block the first HTTP/1 load and start with loading with HTTP/2 first and flush cache between loads, the speedup vanishes. The test is disingenuous, it is not testing HTTP/2 but DNS cache velocity.

So, those type of website makes me rather cautious. And the test, for the small scale workloads I work with, have not been very conclusive.

Do you have serious articles on the matter to recommend ? Preferably not CDN provider trying to sell me there stuff.

The demos I’ve seen use different domain names for the HTTP/1 and HTTP/2 tests. This makes sense, because how else would you make one set of resources load with HTTP/1 and the other with HTTP/2? This deflates your DNS caching theory.

I didn’t rely on tests by CDNs, though. I measured my own website! Accept no substitute! The differences are most dramatic over poor network connections and increase with the number of assets. I had the “privilege” of using a high-RTT, high congestion (high packet loss) satellite connection earlier this year and difference is bigger.

What I like about it is that I feel like I have more freedom from CDNs and complicated tooling. Instead of using a complicated JS/CSS bundling pipeline, I can just use a bunch of <script>/<link>/"@import/import". Instead of relying on a CDN for large assets like JS libraries or fonts, I can just host them on the same server, because it’s less hassle with HTTP/2. If anything, I feel like HTTP/2 makes it easier to make a self-sufficient site.

Finally, HTTP/2 is so dead-simple to set up on your own server, most of the time. It’s a simple config setting.

Are you actually seeing good results from push? I have seen many projects try to use it, but am not aware of any that have ended up keeping it.

(Disclosure: I work at Google)

Push isn’t worth it, from what I understand. I think that’s the conclusion at work.

I _think_ it's working pretty well as far as I can tell.