Hacker News Reader

Comment by vander_elst | Hacker News Reader

It's like 0.13qps on average. Assuming a constant load and no spikes, a raspberry pi should also be able to handle this load, it would be nice to compare the performance

View on HN

My Raspiberry PI model 3 would do roughly[1] ~250 req/s with an Elixir server I wrote. 350k @ $3 requests seems kind of expensive to me to be honest.

Is your pi at an AWS data center? How much to co-host it at an AWS-grade data center?

Sure a self-hosted pi is cheaper, but that's an apples vs oranges comparison.

What leads you to believe it's an apples-to-oranges comparison? From OP's example the AWS monthly bill would be close to 10% of the total hardware cost to serve the same volume of requests.

How about hardware problems if they happen? Or just rebooting it sometimes to apply kernel patches? Multiple fiber links for a case of bad network weather? DDoS mitigation?

This is what you pay for in the datacenter. The request handling itself is much simpler.

The question wasn't about a PI in a data center was it? It was about how does this compare to a raspberry pi.

I'd even question whether or not the application needs to be in an "AWS-grade datacenter" since it does 1 query every 10 seconds but sure...

I'd be more than willing to bet hertzner's cheapest VPS (~$3.25?) would do over a million[1] requests per month pretty easily and they're inside a data center which is a pretty apples to apples comparison.

My point is that the cost per request is high, regardless of data center or not. That's a lot of money for such a low volume of requests.

[1] Which is still only a request like every 2 seconds.

I don't understand why you're getting downvoted. If you're serving simple content, the bottleneck is going to be the network card or disk, every time. In this case, you have static content: a well-written web server (not nginx, but mongoose or similar) can easily handle enough requests to saturate a gigabit uplink on a first gen Raspberry Pi.

Not that it has a gigabit uplink. But handling 100k req/s is trivial for a well-written server.

You cannot ignore infrastructure while saying X is cheaper. AWS Lambda is a "batteries included" solution vs a Pi being "some assembly required".

Hertzner is closer, but even then, it's not AWS Lambda. Setup and maintenance of a VPS is not "free".

Reply on news.ycombinator.com