codinghorror

Think of Discourse as a JavaScript app you load on your phone. Yes there is that unavoidable first time install on first page load, but once the app is installed, all it needs to render the page is the JSON data that makes up the comments -- which is very efficient, far more efficient than a full HTML page load of a traditional website.

So how you feel about this depends on "how often you use the app". If you only plan to visit that one page, ever, then it's not a great tradeoff (though 700kb -- no idea where you are getting 4MB from -- of images is basically nothing on a webpage these days). If you visit many discussions and many pages, it's a win.

In other words a single discussion doesn't need 4000 comments, you just need to read 4000 comments across all discussions all time. I'm not sure your math is correct here, either: I just visited meta.discourse.org in incognito mode and I see about.. 700kb of javascript?

http://imgur.com/a/gPiEb

I'm sorry but I don't know where you are getting 4 megabytes out of 700kb. See above screenshot.

Answering Macha, below:

- Your chosen example is missing http/2 so requests aren't batched

- Discourse homepage by default shows user avatars for each participant, up to 5 in each topic, so there will be "more images"

I suggest testing with meta.discourse.org (hosted) or discourse.codinghorror.com (self-hosted) to make sure you're looking at apples to apples, and properly configured sites.

In my opinion, it is a truism in software that it takes three years to build the thing you originally set out to build. This was true of Stack Overflow, and it's also true of Discourse.

We're supposed to be building software with the users, while observing them, while participating alongside. This takes a few years to get right.

No plan survives contact with users, and that's as it should be.

We have hundreds of live Discourse sites running the $10/month self supported Digital Ocean (or whatever cloud you like) version for literally years without any issue at all: https://github.com/discourse/discourse/blob/master/docs/INST...

I used to do all these installs for people by hand, and personally support them to ensure that we had indeed built something that was easy to support long term. I stopped doing this after a few years because I knew we had.

yes, but the minute you click a _second_ page, the Javascript app is already loaded, and we request only JSON for the new data.

Also this 4MB number isn't correct; I count 700kb. Feel free to double check my work:

http://imgur.com/a/gPiEb

As I like to say, if nobody hates what you're doing, it isn't very interesting.

simply press CTRL+F twice. Note that on small topics, CTRL+F will work as you expect as the entire topic fits on one "page"

yeah this is the one! A $1100 chip for $500, by Intel's math anyway.

not mentioning ECC is pretty sloppy, granted. And virtualization.

The multi-core angle is the main "programmer" bit of Ryzen. If you are writing heavily multithreaded code you will get 2x the cores for your $.

The watt budget of a laptop is prohibitive. The watt budget gloves only come off with a desktop.

True! Basically 2x the cores Intel will sell you, at the same price. IPC is "only" at broadwell levels but that's still pretty solid.

Wait so when someone typoes cnn.com as con.com, that is ipso facto a memory error? I guess I could see that if the characters are radically far apart on the keyboard? But doesn't a simpler explanation like "one person out of billions with Internet access typed the wrong thing" seem a lot more likely?

While I agree CPU perf improvements have been quite incremental, PCI express SSDs and 64gb ram make Hackintoshes worth it.

I almost wish Apple would just quietly make this easier for folks, blessing it without explicitly blessing it.

This is so depressing. Shame on you, Uber.

Lack of http/2 on haproxy by now is terribly limiting and very disappointing.

We don't see many NodeBB sites in the wild. If you want an old-schoool, traditional PHP / MySQL system, XenForo is a good choice, but there are a lot of critical security exploits, so stay up to date.

Only 1Gb is required, which is $5/month on some VPS providers. Redis, Postgres, Sidekiq, etc all need to be running. It's a modern stack.

I totally get the concern as Disqus is a third party with their own monetization and ad goals. But it still seems a bit abusive of GitHub as a free service?

Why not stand up a Discourse instance for your comments? https://meta.discourse.org/t/embedding-discourse-comments-vi...

We find that http://browserbench.org/Speedometer/ -- which is what v8 is optimizing against now -- is extremely close to real world Discourse (EmberJS, Glimmer) performance.

We always wondered why Android was so crazy ridiculously slow at our real world JS vs. iOS starting in late 2013 onward.. getting worse and worse over 2014 and 2015.. and "gee, we optimized for the wrong thing aka Octane" was a big part of that.

http://benediktmeurer.de/2016/12/16/the-truth-about-traditio...

You can read more about the history at https://meta.discourse.org/t/the-state-of-javascript-on-andr...

I disagree. I think the higher levels are many orders of magnitude more difficult.

My heart goes out to you. That is a heavy thing to bear witness to.

We don't get a lot of death intruding in our lives when we're young. But it starts to creep in as you get older and it is ... very unsettling.

Purley is the big Skylake based server platform chipset reset for Xeon E5 and higher so Google must have early access to it.

Skylake Xeon E3 is on the current older server chipsets..

Purley is supposed to be a significant improvement.

I think it's AMAZING that a company offering nothing but scheduling posts on social media got to $13 million per year in revenue. Amazing. And kind of weird.

Images are meticulously optimized, problem is, retina is expensive in file size.

I agree with this. Chat and threading are oil and water to some degree. To me it is the difference between short term and long term memory; in short term memory, forgetting is a feature -- not a design problem to be solved.

Yeah that is a logical way to look at it, and better than $ per core.

Oh yeah those are amazing. Fanless is the holy grail but it is very risky and super hard. See also related review: http://www.silentpcreview.com/NoFan_CR-80EH_CS-60

Excellent retrospective. I know these are painful to write, so thank you.

You are correct the shipped game absolutely used hardware 3D acceleration .. by 1998 3D cards were quite popular, that was the early nvidia and late 3dfx era.

Work somewhere that public output is part of how you're compensated and promoted. See: https://stackoverflow.blog/2011/07/how-much-should-you-pay-d... "One important principle of Stack [Overflow] is that we do as much as we can publicly, and we try to leave public artifacts of all the work we do."