Ask HN: Convince Me of HTTP 2 | Hacker News Reader

Ask HN: Convince Me of HTTP 2

14 pointselgatito32 comments

Possibly a clickbaity title :) but it's an honest question on the hype on the new HTTP versions.

I am familiar with the hypothetical advantages and with the synthetic benchmarks, but does this really translate into noticable, real-world differences in browsing experience?

IMHO it does not. I've had my browser configured for HTTP 1.1 now for months and enabled HTTP 2 and 3 for the past few days, but cannot say there was mentioned noticable difference in how fast sites were loading.

Did I miss anything? Did you notice a significant speedup when you switched to HTTP 2? Please only your real-world experience, as mentioned I know it can have better performance in synthetic benchmarks.

View on HN

> Please only your real-world experience,

Hard NACK, absolutely not.

Noticeable is not relevant. If there are a billion people who load 40 pages and each saves 2ms, we save 2.5 human years a day.

Http2 isn't even the good stuff. It still has horrible head of line blocking.

Where these protocols, http3 in particular, has real greatness is in it's ability to handle lossy & jittery connections. On http1 or http2 with a dodgy mobile connection, recovery from a missed packet or three can be egregiously painful. With http3 the bits can keep flowing.

In general I have high skepticism that you want to be convinced. This narrow on-guiderials "anecdote only" line of inquiry speaks to hang ups, about resistance to allowing progress. You are looking for personal annecodotes, talking about your experience, but I highly doubt humans have the mind to really be able to tell about very fine differences, but that doesn't invalidate time saved, as we load so many resources a year. Last, you likely have very fine internet connections where many of the overwhelmingly grand advantages don't come out.

I am afraid I am not sure what you are saying.

Noticeable is relevant. Either it is noticable faster or it isn't. In my experience it isn't, hence the question.

If you can convince me with a site where one can actually notice that, I'd be more than happy to change my stance, but multiplying 2 millisecond by X humans or using a kbps connection as baseline won't cut it I am afraid :)

As for your assumption on the Internet connection, we are talking about a mobile connection, so not exactly fiber :)

if caring about performance or the general experience of everyone who doesnt have wired gigabit ethernet isn't of concern to you, then you're not going to be convinced.

Who ever mentioned wired gigabit Ethernet? Please elaborate.

My discussion about packet loss & less than ideal connections.

Yeah, the discussion was never about acoustic couplers nor about gigabit connections. And on "regular" connections, the newer HTTP versions do not perform better than HTTP 1.1.

But again, why are you replying if we already agreed we do not need to continue the conversation?

The new https so go faster and you are wrong.

I struggle to think of how you are trying to actively sabotage the conversation & blow smoke this time. Acoustic couplers? Get out. You aren't fooling anyone. You are not this dumb; you are sabotaging the conversation again.

Packet loss is a fact of life in a huge part of the world. Anyone on mobile has some packet loss they contend with. These users deserve good protocols that work. You seem unable & unwilling to acknowledge even the possibility of that. It's unclear why you have such a huge hostile bias against modern improvements.

> If there are a billion people who load 40 pages and each saves 2ms, we save 2.5 human years a day.

That's as idiotic as saying "If a billion people take one less breath a day, humanity will consume 500,000,000,000 ml less air a day."

It doesn't mean anything.

I'll remember this for the next time I try to convince someone that performance is important. However I think it's like saying: it takes one man one day to build a ten meter long wall. If everyone on earth would help building the wall, it would take less then one hour to build a wall around the entire earth.

> we save 2.5 human years a day.

That is such a b.s. metric! What exactly does that 2.5year translate to? Absolutley nothing! If not a single person can take advantage of it then you're just doing a bunch of busy work.

> but that doesn't invalidate time saved, as we load so many resources a year

Yes it does. I think what you meant is how some $bigcorp will save money,but even then , how?

I also dislike how you attempted to interpret OP's intentions. Just freaking give a real world example and that'll shut up anyone who can argue against your view point.

Yes, I do notice a significant speed up with HTTP/3 compared to 1.1.

But I have spent some years of my life playing eSports shooters competitively and I can tell the difference between something happening in 150ms or 250ms and monitors below 120Hz make me uncomfortable. Some people, on the other hand, claim that cloud gaming has no perceivable input lag. So it depends on the person.

With that said, 1.1->3 should be perceivable for most people, I think. Head-of-line blocking is slowing things down significantly. HTTP/3 latency feels like browsing with UDP, which it is.

Of course, browsers do caching and pre-emptive loading, so the user experience in many cases will be the same. But when loading a website for the first time, or reading an article with many images for the first time, I think the difference in page load/draw times can be easily observed.

EDIT: On further investigation, it seems like 1.1->3 should be perceivable for most people if they are on poor quality connections, and if the website/web app is heavy. Looks like I fall into that bracket with some of my devices as I live in an ancient house with stone walls, the natural enemy of WiFi.

Would you have a site where that can be reproduced?

While I am not saying it might be to some extent subjective, I cannot say I have noticed anything noticably different between and 1.1 and 2 or 3. For that reason I'd be curious on which site that would be noticable.

YouTube, for example. I haven’t done a rigorous test and looked at the timings. But I disabled HTTP/3 in my Safari settings just now and browsed around YouTube in Private mode. Then I did the same with HTTP/3 enabled. It was at least 3x faster at initial page draws with HTTP/3. I did this test just now on weak WiFi.

Thanks for the example. Though, Youtube was definitely one of the sites I visited.

I just ran a test with HTTP 1.1 and HTTP 3 - and in both cases, it took about 10 seconds until the whole page was loaded. It was already usable earlier than that, but even that was similar between 1.1 and 3.

The test was done on a "regular" Wifi connection.

I just tested how long it took to get full content draws of YouTube over HTTP/1.1 and HTTP/3 with my wired 1Gbps machine and Chromium-based Edge (with InPrivate and no extensions, obviously). I did three tests a couple of minutes apart.

Timings:

HTTP/1.1: 2.54s, 2.18s, 2.51s

HTTP/3: 2.07s, 1.95s, 2.40s

What does N=3 mean? Probably not much. The findings confirm my feeling that HTTP/3 is faster, but it is difficult to separate signal from the noise with so few readings. Plus, reasonably, I understand that other factors can bias these readings much more.

Moreover, I don't think that this matters as much in wired connections where packet loss is very minimal compared to WiFi, especially bad WiFi. Over a weak WiFi connection, subjectively, the full content draw timings felt significantly different to me. Though I didn't do a strict measurement. I wonder what these timings would look on cellular.

Thanks for the test, these rather very close values reflect my experience as well.

While my Internet connection is actually based on a mobile connection, it's still difficult for me to comment how a classic smartphone may behave, as that's not my typical use-case. Here, UDP may offer some advantages, if the connection quality can vary.

I wonder if it was a fair summary to say, that HTTP 2+ (and especially HTTP 3) may offer advantages for unreliable connections, but not so much over somewhat reliable connections.

I don't believe this for a second. When you play a game, you know what performance target it should be hitting, because you can visually see it.

There's no meaningful performance target for websites. Instant would be great, and so for matters of debate, let's say that's the next vsync poll and go with the most common poll rate that exists today, ~16ms for 60hz monitors. Is that going to happen? No.

Since you don't know what the up/down speed is for any random given website, there's no way to know if what you're experiencing is just the server with a low upload speed to to your client, or the protocol itself.

And the Internet connection of the website itself, the language the web server is written in, the performance of the software in question, and other factors are going to have dramatically more impact than the protocol version in question.

> There's no meaningful performance target for websites. Instant would be great, and so for matters of debate, let's say that's the next vsync poll and go with the most common poll rate that exists today, ~16ms for 60hz monitors. Is that going to happen? No.

You set an unreasonable arbitrary target and then say neither 1.1 nor 3 will reach it. Yes, I agree. But I don't understand how this is useful.

Cloudflare's QUIC test page gets drawn in about 200ms for me with HTTP/3, while it takes about 250ms to draw it with HTTP 1.1. 50ms is 2-3 frames on a v-synced 60Hz monitor.

I see the same "few frames quicker" effect on many basic websites that support HTTP/3, and much more significant changes on heavy websites or over poor infrastructure where head-of-line blocking can hurt more.

I think any meaningful improvements that could have arrived by improvements to the Hypertext Transfer Protocol itself will have and can have been eaten up by all the stupidity of modern web development.

So no, I don't think anything meaningful has arrived in terms of performance, otherwise we'd all be singing its praises.

Rest In Peace, HTTP/2 Server Push.

Its not noticible at the client side as much it is at the server side. Every millisecond saved at server side allows it to serve other thousands of connections.

IIRC, There are studies which show a difference, especially in poor latency scenarios.

The issue with HTTP/2 is that it complicates the stack: ideally one would support either HTTP/2 or HTTP/3 but there's little point in supporting both, since you always have HTTP 1.1 to fall back to. HTTP/3 offers best performance, HTTP/1 best compatibility, HTTP/2... doesn't offer that much in comparison.

Thanks, but did you personally experience noticable better request times with the newer versions?

I couldn't make the comparison since I upgraded my internet connection at the same time. I strongly suspect most (all?) of the improvement came to that.

Genuine question - why is it up to us to convince you? It's already in widespread use benefiting the folks who use it.

No. Instead I will convince you in opposition to HTTP.

Switching from HTTP to WebSockets across the board increased the test automation in my app 7x. This also simplified a massive amount of administration and maintenance in my application regarding micro services.

HTTP2 is better than HTTP for downloading many assets simultaneously, but if you want your web application to be fast you need to bundle assets as much as possible there are fewer things to need in parallel.

My question is, why not? I think we can all agree it's better, whether or not a noticeable amount for a given person doing a given activity, and it's pretty well baked into the stack at this point

I would say server push is the main selling feature. It allows you send the assets together with the first request in order to save a few roundtrips. It however took years for the major browser to implement it properly eg. Not re-downloading the pushes files. And now server push is being deprecated.

Take a flight and use the onboard wifi to visit websites.

Reply on news.ycombinator.com