Hacker News Reader

Comment by TheIronMark | Hacker News Reader

> DIYing it does come with additional maintenance burden, but so far we’ve found this manageable.

Tune in next month for their article on how much fun manually scaling everything is.

Bare-metal is so cheap that you can turn everything to 11 to begin with, never have to worry about "scaling" and still come out ahead in terms of cost.

In fact, before the cloud craze came around, internet services used to run just fine despite autoscaling not being commonplace. For a lot of use-cases, autoscaling is a problem created by the cloud's high prices and is unnecessary if hardware is cheap enough to run continuously.

The way I look at it is that the cloud doesn't run on the cloud. It's totally artificial.

That's only true if you don't factor people hours into your costs.

The vast majority of companies have one if not more DevOps/infrastructure engineers on payroll. The "people costs" only realistically works either for super-early stage companies where the founding engineer(s) also manage the infra or something that's simple and standard enough that it can run on a fully-managed PaaS such as Heroku.

But hey if you're gonna bring up people costs, shouldn't the simpler stack (because it doesn't need to autoscale) mean less management & maintenance overhead anyway?

The reality is that most startups don't scale exponentially unless they are injecting huge amounts of cash into their company (and in that case they can afford AWS). The company in this article just has a single server.

However, most services that are around for a long time experience bursty traffic so provisioning is still difficult.

Fun fact: Hacker News runs on a single core.

Sarcasm?

"HN's main software runs on a single core", wrote a moderator when I wrote to ask about why rapid (<2s?) subsequent actions were rejected by the server (ie. like a post and then favorite it).

This was November 2020 and he mentioned a big overhaul was in the works (of both HN and Arc, the programming language in which it is written).

We've seen new features since then, but the throttling is still there so it seems to still be the same situation regarding fundamental infrastructure.

I didn't know that. thats very impressive.

It really depends on a lot of factors such as technology, amount of traffic, etc. My server handles about 200 simultaneous users on a few % of cpu. I think most startups will have a smaller load than this unless they become incredibly successful or pour tons of cash into growth.

Reply on news.ycombinator.com