Moving away from AWS reduced our annual server costs by 80% – from $1M to $200k | Hacker News Reader

Moving away from AWS reduced our annual server costs by 80% – from $1M to $200k

levelup.gitconnected.com

97 pointsdlcmh20 comments

What technology stack do you use, and why did you choose this stack?

We use Javascript everywhere, since we solve the “issues” caused by Javascript rendering we want to build as much expertise as possible in this field. But for the other parts, we are taking advantage of CloudFlare’s distributed system for fast response and global scalability. While our uptime guarantees are supported by Digital Ocean’s cloud platform. We also use a myriad of other SaaS providers to maximize our effectiveness.

Quoting in case you were looking for this part like me.

I'm not 100% sure what the destination platform was that was only 20% the AWS cost. Digital Ocean? A "myriad of other SaaS platforms"? Self-hosted in a colo datacenter somewhere? How does this newer solution scale? What are the intangible costs, and are there additional staff considerations? Do you answer the pager at 2AM when a physical piece of hardware goes down?

Digital ocean doesn't charge traffic costs. I'm not sure if that was used in the article, but DO can provide significant savings for high-bandwidth services.

Since when? When I used them they charged $10 per TB of bandwidth

You now get the first TB for free per droplet. So if you have 5 droplets thats 5 free TBs outbound per month. Inbound is always free.

https://docs.digitalocean.com/products/billing/bandwidth/

They're also part of the bandwidth alliance, unlike AWS.

Digitalocean is not a member of the bandwidth alliance

Ooh, surprising. I thought I saw their logo in the wall of logos last time I checked.

First paragraph says "in-house infrastructure"

Previously: https://news.ycombinator.com/item?id=33077876

One powerful reason to use aws is autoscaling. Everything else is bs.

But you can use your own hardware for most of the traffic and use AWS when you have spikes.

How would you go about doing that?

Really basic version, K8s cluster with the fronting off of AWS where AWS compute joins the cluster as part of the scaling portion.

Can also use fancier concepts like tiered offloading on your non-AWS to S3 if you exceed say the bare metal disks because of some giant wave of ingestion. Apache Pulsar has that built in.

An ALB fronting your service pointing to a service pool that has only your hardware at idle. In the event of a sudden growth of popularity, add AWS hosted workers to that pool to spread the load.

Disk and snapshot costs aren't bad either which leaves me to use AWS as only a backup solution as it imposes no outbound traffic.

It's easy to move if you don't use custom functionality of a particular cloud provider. That seems to be the case in this story.

It is not mentioned in the article but it seems they have moved to Hetzner, which has free ingress/egress.

https://dnslytics.com/ip/5.9.82.52

Hetzner is a too good to be true service. Only that they have less locations but I'd imagine they need a very well controlled data center and staff to reach that kind of cost and can't expand easily.

This has been on hn previously but can't find the threads.

Reply on news.ycombinator.com