Hacker News Reader

Comment by idlewords | Hacker News Reader

It amazes me how many people still can't configure a webserver in 2023 to not fall down under a few queries/second load serving up a textfile.

View on HN

Is the HN hug of death truly only single digit QPS?

I've had posts on the front page of HN (#1 or #2). That generated about 20k visits.

Assuming:

* 1 page/visit, which is the modal value based on my analytics

* the period where it is on the front page is 6 hours (I don't know this, but I could if I could be bothered to go back and look; I don't have granularity down to the minute, though)

That is 20000 requests/(66060), which rounds to 1 request/second.

Of course, there are all the other assets (JS, images, etc) that you need to account for as well. And it might spike to higher.

I think that's completely useless? The whole point of qps is that it gives insight into spikes.

If the entire page is 50 KB, it think within five minutes, it can be served 20,000 times over typical commercial broadband (assuming 30 Mbps net capacity)? 20,000 requests can't really form a huge spike. Most personal-ish sites won't be hosted on commercial broadband, so they should have more bandwidth than that.

That assumes the hits are evenly distributed.

I think it’s more likely that there’s a big flood at first and then it drops off towards a low number.

That's been my experience. But even the tiniest pizzabox server should have no trouble at 40+ qps.

Genuine question, what is it causes websites to keel over in a mild breeze like that? Surely not raw CPU, memory, or network bandwidth on the machine? Is it just some arbitrary limit imposed on whatever VM they are running?

> Surely not raw CPU, memory, or network bandwidth on the machine?

Why not? It gets worse and worse as we aim for more abstractions and layers of tooling.

Where the latest hype is server rendering React in NodeJs for what instead code by a static site...

Reply on news.ycombinator.com