It's quite cheap to run. It's just one small EC2 instance, and it's usually idling around a load average of 0.1 (though right now it's at 1, hehe). It costs me about $24/month. The Apple developer license costs $100 per year, so it's ~$400/year. As of recently, that's entirely covered by donations.

I am incredibly humbled by the sponsorships. I would have never thought ntfy would take off like that. I love open source and I promise it'll always be free and open (as long as I can reasonably fight of the abusers).

If you'd like to help me out me, you can donate via GitHub Sponsors: https://github.com/sponsors/binwiederhier

If you would like to lower the cost more, I think switching to a cheaper cloud service like Linode or Hetzner would be the way to go.

I'm not sure if it's worth going any lower than $24/month

If I were to run my own personal instance of this that cost would be quite high. A free Oracle Cloud VM or $5 Digital Ocean droplet would be the way I go.

A free Oracle Cloud VM still costs you your soul, in perpetuity.

That can't be right, some of us have worked for IBM and no longer have souls.

IBM and Oracle have a soul-sharing deal signed.

They went to court over it and depending on the number of CPUs you use you might be forced to have one or two kids to pay the extra souls.

Yeah, I think that "or two" in "one or two kids" is still being litigated. My reading is that it'll stay in litigation indefinitely, as this is the price both companies agreed to pay when they decided to retain lawyers from Hell itself.

Well, of course. “$24 per month seems high for running my own personal instance that’s going to do a handful of notifications daily” vs “$24 per month seems high for running a public service that could very easily need to be scaled up quickly” are two very different statements.

If I go by $50/hour salary on average for developer, if someone saves 20 minutes per month on average using AWS, it is worth it.

Except where worth it in dollars < worth it in hassle

If it's a single EC2 instance and not using any other AWS services, I fail to see the difference between Hetzner, DigitalOcean or AWS.

That being said, it's probably not worth migrating to another provider at this point.

I only have experience with Digital ocean and AWS. AWS free support is definitely much better than Digital Ocean. Also Digital Ocean is more quicker to delete your data than AWS in case of billing issue or anything.

Would the $5 droplet scale (and charge appropriately) if their specified bandwidth is exceeded?

I mean from 24 to 10 is a big difference.

You can probably count down server cost by half. Have you tried with lower resources or have figured out a minimum requirement for the server? As an experiment I suggest - https://tinykvm.com/ - with FreeBSD or OpenBSD (Linux doesn't really do well with small amount of RAM).

Just gave the developer $50 so he doesn't need to reduce monthly costs. I like the service running well.

Much appreciated my friend. I honestly didn't opt for the cheapest server because I wanted it to run well. I don't want to constantly fight for resources or worry about it. It was supposed to be fast and be able to handle traffic well.

And it is. It's not falling apart from the HN traffic and it has a lot of head room. Plus. It's doing 400k messages a day already easily.

Anyway. Thank you so much for your generosity. I am humbled and thankful for your support.

Your approach is great. It’s odd to see so many penny pinching comments here, the HN crowd usually skews the other way.

I think its because this is a hobby project and for a lot of people the cheaper a hobby project is the more they can keep running before the money runs out. It's also that for some people optimizing costs and making things run on as few resources as possible is fun.

But if you don't find it fun spending time performance optimizing to the extreme, and can afford it, then there's nothing quite like massively over-provisioning hardware. It makes a lot of performance and reliability problems go away.

Probably some truth to all you said. I spend way more money on my hobbies.

For this creator, his hobby is well into the realm of "could productize some component", even if just creating derivatives of this for some commercial purpose. I'd much rather spend time on that than the optimization part. My thought is HN has skewed more technical than entrepreneur over the years and the folks who are into the optimizations are more vocal in the thread. Could also just be the cohort of HN that was attracted to this article. "Send push notifications to your phone" sounds pretty technical. Or, the tech recession is changing mood about money and I'm not in tech.

That is nice of you and you've helped them to cover server costs for the next two months. Server optimisations can help them reduce the cost further.

It is super nice, isn't it? I am so amazed how awesome people in the open source community are.

As for the scale question: The server is already tuned for scale: https://ntfy.sh/docs/config/#tuning-for-scale

If you have any more ideas of what nobs to turn, let me know.

I appreciate the dogged pursuit of efficiency, but spending a lot of time optimizing for the current scale and getting costs from $24 per month to $10 per month (or whatever) is likely wasted time if this grows significantly. You have to be able to recognize there’s a lower bound where your costs are low enough it doesn’t matter and even saving 75% isn’t worth thr effort. The author has obviously drawn that line somewhere above $24 per month and that’s perfectly reasonable.

That's a good point. I guess many of us were suggesting cheaper VPS because we assumed this is a new project, just launched. (Most inexperienced developers / start-ups tend to over provision). With current hosting technologies, it is quite easy to scale with need, with minimum downtime, so it makes sense to start with the minimum server requirement and then go for a powerful server as required. This give you valuable data on the bottlenecks (where you can do future optimizations), as you scale, while you temporarily fix the issue with more money (by requisitioning better servers) as you work out the cost-benefit analysis (as you did) of spending money on optimising vs server upgrade.

This is super cool. If you don’t mind saying, how many connections do you tend to keep open at steady state?

Here are some stats from last night showing the volume of messages:

   IPs: 9756
   Clients: 9980
   - Google Play: 2796
   - F-Droid: 5985
   - iOS: 794
   - curl: 62
   - other: 343
   Messages: 
   - Successful: 310076
   - Failed (rate limited): 104741
   - Failed (other): 6511

   Top publishers (* = limit exemption):
   22908 matrix.org*
   22352 matrix.gateway.unifiedpush.org*
   18286 up.schildi.chat*
   16134 192.124.x.x
   14056 matrix.envs.net*
   13376 108.35.x.x
   7794 114.132.x.x
   6035 112.78.x.x
   5432 88.99.x.x
   5366 138.201.x.x   

In terms of open connections, it varies, but right now it's 5k active connections. But there are many many self-hosters out there, and the people using Firebase (no "instant delivery") are also not counted in that.

You can subscribe to the stats above using the https://ntfy.sh/stats topic (old messages are only cached for 12h, so half the day it shows up as empty)

Watch out for data transfer costs on aws after HN

Thanks. I'll keep an eye on it. I have a lot of rate limiting in place, but for static resources it could probably be tighter.

Thanks this is really nice. Impressive to see that f-droid is that high!

Hey! nice idea and software. With the costs you mention how many notifications can you send? There must be a limit right?

I love your work and admire your approach. But you should recognize a void will emerge and void get filled. Someone (probably reading this thread) will clone your idea and make it a supported for pay service.

It wont be me. But some enterprising engineer/pm/group who was recently laid off in this downturn will probably make a go of it.