Hacker News Reader

Hacker News

Comment by TylerLives | Hacker News Reader

Full thread

TylerLives

How much does it cost to run a website like HN?

View on HN

A lot using redis as the backend.

Why would using Redis add significantly to the cost?

Because Redis' database is stored exclusively in RAM, which is an order of magnitude more expensive than even SSD.

Also, long-term maintenance effort and costs would be substantially higher. (Redis persistent-to-disk options, namely AOF and RDB, are notoriously unreliable.) These issues would persist (ha) even with a clustered setup (actually, especially then.)

I've never heard of them being unreliable, and certainly not notoriously so. What sources do you have on this? My own impression formed from reading HN is that opinion around Redis is nearly entirely (almost cultish-ly) positive.

I'm just checking if there is something I missed.

> Because Redis' database is stored exclusively in RAM, which is an order of magnitude more expensive than even SSD.

Maybe, but still very cheap. Unless you are talking about insanely high priced cloud services.

Something like Hetzner would provide you with 128GB of ram for 60 euros / month or 1TB of ram for around 500 euros / month.

Not exactly true. There are things like "Redis on Flash" that redis.com offers for large sets of data. If you enable it, most of your data gets stored in a Flash memory and hence super cheap, in fact even cheaper than things like DynamoDB and stuff.

Probably surprisingly little. Also, the publicity HN brings is most likely orders of magnitude more valuable and that makes it much less important to try and optimise the service.

The pages are very light and the algorithms seem to be designed to be quite light, too.

Now, I know neither Arc nor the traffic stats, but I would not be surprised if one or two machines were able to serve entire traffic.

I don't know how much traffic HN is getting, but i think you should be fine on a 10$ VPS for the most part

2018 is was around 6m requests/day https://news.ycombinator.com/item?id=26469566 Hardware used at 4m/day was two machines but it runs on one machine https://news.ycombinator.com/item?id=16076041 So maybe double that now.

(update: I mixed up M5 hosting company with AWS M5 instance types)

Do you have to look at the peaks and design for those? (The hardware link accounts for this of course, but not sure the hardware scales linearly with average load, because the peaks may not scale linearly with average load)

Related question: is it possible to run HN and serve HN-level traffic on any of the major cloud providers' free tier?

Maybe Oracle Cloud. They give you up to 4 cores ARM64 instances with up to 24GB total RAM.

Reply on news.ycombinator.com