Hacker News Reader

Comment by joshstrange | Hacker News Reader

This article left a really bad taste in my mouth. I don't believe GoSnaps == GoChat in terms of complexity and the constant back patting and self congratulating is really distracting. There were a couple of decent takeaways but largely the whole post revolved around how "How smart am I?" and "What great foresight I have".

I really don't approve at all of the GoChat shaming going on. The author may be 100% correct that GoChat made mistakes in writing code that doesn't scale well but that doesn't give him a blank check to beat and berate GoChat. It reads as a very discouraging post to newer/less experienced programmers in my opinion, essentially "Don't even bother making something unless you know it can scale to millions of users" which I think is a terrible message to be sending.

View on HN

The real take away is that experience beats inexperience. GoSnaps author seems to have more experience and GoChat author seems not to be a developer and outsourced his service. Without prior experience, it's difficult to have insights into all the things that could possibly go wrong.

I personally don't feel like he was berating GoChat, if anything I was sad for GoChat to have grown and fallen under it's own weight, but yet still enough respect for GoChat to have had the foresight to build and execute. The author even called it a "genius move"

> I personally don't feel like he was berating GoChat

You don't feel that way because he edited his article and in its original form it was VERY uncharitable to GoChat. Now, after the edits, it's only lying by claiming they're similar in performance profile, when in reality, in terms of realtime performance, GoChat is all GoSnap does PLUS a lot more extra work.

And "oh, look, they spend $4,000 a month on servers, and I only spent $100/month on a server. Plus 'cheap' storage from Google, that I'm not going to enumerate for some reason"

A screenshot on my iPhone 6S of Pokemon Go (just the map) is 3.7MB, if we extrapolate we are talking about 722.66 GB of images so far (using his 200K images estimate) we are talking about < $20/month for the storage and ~$80/mo for the bandwidth (Sharing them back out, ingress is free). Now this is early days for GoSnaps so it's hard to predict if those users will stay but to the author's own point he has to keep all of the pictures so that number will only grow. It is disingenuous to completely leave out a cost of the service which looks like it has already probably doubled his cost per month from his "Only $100 for a server" claim. Also I'll point out they may be compressing/downsizing the hell out of those photos but I can't tell how much each one will weigh without MitM-ing the app and I don't care that much.

For sure - I certainly wouldn't expect that it'd come close in hosting costs, but when the author is as snarky as he was about how cheap, then he should expect to be called out.

And besides (leaving aside the numerous other 'issues' with the claim, as well discussed here) is "My $200/mo stack performs better than a $4000/mo stack" really that much worse than $100?

I agree. The complexity of his photo app is nowhere close to something like GoChat.

Plus the condescending tone when he is just copy pasting code from other project is really annoying to read. Apart from this the content is interesting though.

> condescending

This is exactly the word I was racking my brain for! I love reading about how to make something more scalable by there was so much filth I had to wade through to get to that in this article...

The only points I can pick out are: use node, make sure your db queries are on one index, and make sure you turn off features of modules/frameworks you're not utilizing.

A lot of condescension for 3 small points.

I upvoted because I'm just now hearing about the whole GoChat story, and because I find amusing how quickly Pokemon Go related apps can blow up right now, but yeah, it is evident that this was primarily a self-promotional post otherwise.

Yeah it's complete bs. I've worked on chat before and even 100 clients destroy a core or even two.

Now I would use pusher but that's extremely expensive but scalable.

100 clients destroying a core or two? Please, use the right tool. See erlang, golang, java, even c#.

Years ago, you had to suffer using poll(), select(), whilst dealing with threads or forking. It's so easy these days. A raspberry pi should be able to support 1,000 clients with any of the above recommended languages.

That seems unnecessary. Or did you use (something like) Meteor.js?

Yeah, no, that's wrong.

If you have problems with this, there's something really wrong with either your design or your choice of tools.

How'd you go about building this?

In 2008 with php and ajax

Node was designed to hit 100K users/core, and in fact can do closer to 1M users... for simpler chat relay, a single server should be able to handle 1M simultaneous users... Growing beyond that means you need to segment/orchestrate your channels. I haven't used the app to comment... but something written in golang could probably handle a couple million users on a single server via websockets (or raw sockets), with appropriate backing services in place.

The original article is condescending to say the least, that doesn't mean that $4k of ongoing hosting costs shouldn't be able to handle a few million simultaneous users... though, something tells me that Pokemon Go could approach twitter or FB overhead, and related apps like chat could see huge swarms as a side effect.

That kind of problem would kill most first generation apps, as those who've seen twitter grow can attest to. I don't think there are any IRC networks that big at this point.

Reply on news.ycombinator.com