Why I run 5 different applications on the same server

Sounds awesome, ping me when you have the writeup, interested to learn about your setup.

Here’s an article I wrote about my setup:

Robust NodeJS Deployment Architecture

Same here! And good thing, after 7 years building apps that haven't made money, it'd be crazy to pay for a separate instance for each.

I also really like LXC. I know enough about Docker now that I could probably switch to it, but LXC allows me to work on each app like it's on a separate server, while still giving the root access to do backups and stuff.

I run ~5 PHP apps on a $5 Linode, but there's barely any usage. It can certainly be done.

I strongly believe in keeping my side projects costs to a bare minimum. My only other expense is a handful of domains.

That's cool! I was wondering if 2-4 Gb would be enough to run a Tomcat server, NodeJS, and a database. Do you have a caching tier, e.g. Clouflare? Where do you store your files?

Nice post. I do something like that with http://dokku.io/ , to me is the easy way to use Docker. And I use swap to increase de memory. The best 5 dollars of my life. :D

I always say people overdo their deployments. But I still rather separate projects by VMs personally, it's kind of clean. The big saving is to run your database on the same box, which is even faster (no network trips).

The important thing is to separate the application properly from the rest of the system. I am just writing a book that goes into these bits in detail (rootless services, rootless containers, SELinux). If you have good boundaries, you can easily run everything on the same VM and save $$$.

I ran an Alexa top 50k site off of a single t2.small instance on AWS and never saw the CPU go as high as 10%. People vastly over-estimate how much resources their projects will need.

Did you consider using serverless infrastructure like S3, Lambda, and DynamoDB instead of self hosting everything? I like the pay-per-request pricing of serverless architectures and I wonder if the overhead and cost benefits would outweigh a setup like this.

I'm also using GCP for Newsletterss, but I have Cloud Run instead of a VM, use Pub/Sub for queues, Datastore for the data layer and Go as the main programming language. Firebase Hosting is my CDN and some assets are pulled directly from Cloud Storage (pending to be migrated to a different CDN soon)

It is an unusual setup, but it was optimized for cost reduction from the beginning. I pay less than $1 a month (yup, one dollar) but it's only about 10k requests per day and most of the cost reduction is due to the services being covered by GCP's free tier

I like to roll with either Firebase or serverless usually works well in regards of saving money, as it's pay per use and as long as I don't have much traffic it will be pretty much free.

Everything is automated here as well plus I get the scalability out of the box.

If you can live with the vendor lock-in I think this is even better than rolling your own servers but that's just my 2c.

I'm pretty much the same. I have two separate servers at digital ocean for resilience through a load balancer to make updates and patching easier without taking anything offline, but its just 2 servers that run everything, from websites, databases etc.

If I get enough traffic where it becomes a concern then potentially I'd move a project onto separate servers but not anywhere near that yet.

I do similar for my project. I have a Linode VPS with 2 gb mem ($10/mo) where I have deployed all of my projects. Currently it have 3 web apps and a bunch of client projects mostly cron jobs.

None of my projects get a lot of traffic so its working fine till now.

Well done, balancing costs with productivity and product goals is an important activity for an entrepreneur.

How are you managing your ports? Are you redirecting to different ports depending upon the application or do you have a single service to route to different applications depending upon the paths of a single domain?

What drives costs up, even on a small scale, is data processing with a lot of disk writes and network traffic to fetch data. Responding to user actions on a web app is very cheap when you don't have many users.

I completely agree that minimizing server costs should be one of the higher priorities in the beginning. It really sucks your energy if you're paying a $100 cloud build the first 12 months when you're barely generating any traffic or visitors.

With that said, minimizing server costs does not always equal the most simple server setup. For example, imagine you're building a project that consists of a website and a web crawler.

The optimal solution might be a small instance for the website, but an horizontally scalable solution for the crawler. Suddently we might be looking at a solution that include private networking, kubernetes, containers, auto scaling, multiple instances etc, and this might be the "most cost efficient" solution since it would allow us to increase our costs as the crawler runs, and reduce it when it doesn't, while still keeping the website the same.

Awesome, love this minimalism. If you want push-to-deploy but self-hosted on your own server check out Piku. I am running more than 20 sites on a $10/month Digital Ocean box with this.

What's the reason for hosting nextjs on it over say Vercel?