Low bandwidth users

Continuing the discussion from Make Discourse play nice with the Wayback Machine:

[quote=“codinghorror, post:7, topic:34579, full:true”]

Where are you getting this data from? [/quote]

Measuring from my own site on a text-only topic, no cache, I’m getting 4.3 MB. At 56k simulated speeds in Chrome dev tools, it churns for 15-20 minutes and times out, less than a quarter of it loaded. (I’m still trying to find out why my site is heavier than meta… ) I have yet to successfully load a page at this speed.

I’m not saying there are lots of dial up folks out there. But there are a few left, by choice or necessity.

Verizon inherited about 2.3 million AOL dial up customers, according to Time magazine. And 3% of Americans still use dial up at home, according to this research group.

On top of that, folks throughout the developing world have limited access speed and data caps - not to mention under-powered hardware. A simpler, lighter interface will open the door to them as well.

Obviously we aren’t catering to these folks… but since we already serve up an stripped down HTML version, making it functional will let them participate.


Here’s some other big sites that present a functional stripped down version.

Facebook now has “Facebook Lite”: http://techcrunch.com/2015/06/04/download-facebook-lite/

Gmail has basic HTML: https://mail.google.com/mail/?ui=html

Twitter with javascript disabled redirects to https://mobile.twitter.com, and is functional.

Google Search with javascript disabled redirects to an older but functional version (and slims down from 362kb to 62kb).

Is that without any form of built in HTTP gzip compression? That does not seem correct to me at all. Remember text compresses at 70-80% typically.

2 « J'aime »

Ah! Thanks for that… Turns out gzip was not working. We’re down to about 900kb now, and things are much more snappy!

2 « J'aime »

The irony is that Discourse, being a JavaScript app delivered over HTTP, is a best case performance scenario for a severely bandwidth constrained user. The first load will be the chunky, but all highly compressible, JS text of the Discourse app – and every subsequent request is for only the JSON data necessary to render the requested view.

At no point is a “new web page” send down in its entirety to the user.

But do not take my word for it. Feel free to measure it yourself – with a normal, gzip compatible http connection – and see.

(It is true that first load perf will not be amazing because we send the app in that first payload. But all text.)

4 « J'aime »

You might find this an interesting read:

https://meta.discourse.org/t/blog-post-is-high-bandwidth-required-to-read-a-discourse-forum/26995?u=erlend_sh

(please excuse the slightly aggressive language; it’s not directed at you! At the time of writing, there was a very vocal minority of people who were crusading against the advent of JavaScript as a web technology without presenting any factual data)

4 « J'aime »

After first load, Discourse is an exceptional low bandwidth site. I can’t think of any site that does better in these scenarios.

The first load can be quite rough though, since you need to pull down the whole app bundle. We are looking at improving first load time in various ways in 1.8 or maybe 1.7 depending on timing.

5 « J'aime »

Music to my ears. We have thousands of users in LatAm on spotty 3G connections who find us via search engines, but it’s hard to know how many we lose because they dont wait for the great web app experience to load :slight_smile:

2 « J'aime »

It is definitely quite high up on our list, right under …

:kissing_heart:

4 « J'aime »

6 ans plus tard et je suis curieux d’un point d’entrée léger aussi. Je ne suis pas l’administrateur de notre forum Discourse, mais je fais la plupart du triage des questions de support.

Il y a 3 jours, j’ai atteint mon quota de données mensuel et j’ai été limité. Pendant les 3 jours suivants, j’ai été essentiellement banni du site. L’écran restait sur le premier écran vide avec les 5 points de progression dans une boucle apparemment sans fin.

Heureusement, j’avais activé le mode maillist dans les préférences de mon compte. J’ai donc pu continuer à répondre aux visiteurs.

La question de dépannage ci-dessus concernant l’exécution de gzip est intrigante. Le site a soudainement eu une animation de points de progression il y a quelques mois et le temps d’initialisation est passé de 4 à plus du double.

Existe-t-il un test simple pour déterminer si le serveur livre du contenu gzippé ? Ou quelle quantité de données est transmise ?

1 « J'aime »

Si vous êtes sur notre hébergement, ou si vous utilisez l’installation Docker officielle, gzip et brotli sont tous deux activés dès le départ.

Vous pouvez tester votre site sur https://www.webpagetest.org/ pour confirmer, ou simplement jeter un œil dans les outils de développement de Chrome, Safari ou Firefox.

3 « J'aime »

Peut-être pourrait-il y avoir une entrée d’urgence légère ?

Qui aurait une connexion de compte qui mènerait à un écran de gestion de compte à faible bande passante qui n’aurait que 2 options : activer/désactiver le mode liste de diffusion et envoyer par e-mail un résumé d’activité de la dernière heure/jour.