Cloud services are actually really nice and convenient if you were to ignore the eye watering cost versus DIY.
Specifically, half of the services operate locally, and the other half require CF services. I mainly use Claude Code to develop, and it often struggles to replicate the local environment, so I had to create another worker in CF for my local development.
Initially, the idea was to use CF for my side projects as it's much easier than K8S, but after wrestling with it for a month, decided that it's not really worth investing that much, and I moved back to using K8S with FluxCD instead, even though it's overkill as well.
Ughhhh that is because nobody ever looks at the readme so it hasn't been updated basically since workerd was originally released. Sorry. I should really fix that.
> Specifically, half of the services operate locally, and the other half require CF services.
workerd itself is a runtime for Workers and Durable Objects, but is not intended to provide implementations of other services like KV, D1, etc. Wrangler / miniflare provides implementations of most of these for local testing purposes, but these aren't really meant for production.
But workers + DO alone is enough to do a whole lot of things...
I'm mostly using terminal agents to write and deploy code. I made a silly mistake, not reviewing the code before merging it into main (side project, zero user), and my durable object alarms got into an infinite loop, and I got a $400 bill in an hour. There was no way to set rate limits for AI binding in workers, and I didn't get any notification, so I created a support ticket 2 months ago, which hasn't answered to this date.
That was enough for me to move out of CF as a long-time user (>10 years) and believer (CF is still one of my biggest stocks). In a world where AI writes most of the code, it's scary to have the requirement to deploy to a cloud that doesn't have any way to set rate limits.
I learned the hard way that I must use AI Gateway in this situation, but authentication is harder with it, and agents prefer embedded auth, which makes it pick AI binding over AI Gateway. With K8S, it's not easy to maintain, but at least I can fully control the costs without worrying about cost of experimentation.
10% is the number I ordinarily see, counting for members of staff and adequate DR systems.
If we had paid our IT teams half of what we pay a cloud provider, we would have had better internal processes.
Instead we starved them and the cloud providers successfully weaponised extremely short term thinking against us, now barely anyone has the competence to actually manifest those cost benefits without serious instability.
GCP pricing is absolutely wicked when they charge $120/month for 4vcore 16gb ram, can get around 23 times more performance and 192gb ram for $350/month with Xtbps-ish ddos protection.
I have 2 2x7742 1tb ram each, 3 9950x3ds 192gb ecc, 2 7950x3d's all at <$600/month obv the original hardware cost in the realm of $60k - the epyc cpu's were bought used for around $1k each so not a bad deal, same with ram overall the true cost is <20k. This is entirely for personal use and will last me more than a decade most likely unless there are major gains in efficiency and power cost continues to grow due to AI demand. This also includes 100tb+ hdd of storage and 40tb of nvme storage all connected with 100gbps switch pair for redundancy for a cheap cheap price of $500 for each switch.
I guess I owe some links: (Ignore minecraft focused branding)
https://pufferfish.host/ (also offers colocation)
telegram: @Erikb_9gigsofram direct colocation at datacenter (no middlemen / sales) + good low cost bundle deal
anti-ddos: https://cosmicguard.com/ (might still offer colocation?)
anti-ddos: https://tcpshield.com/
I don't think after the fact that ram prices spiked 4-5x that its gonna be cheaper to self host by 100x, Like hetzner's or ovh's cloud offerings are cheap
Plus you have to put a lot of money and then still pay for something like colocation if you are competing with them
Even if you aren't, I think that the models are different. They are models of monthly subscription whereas in hardware, you have to purchase it.
It would be interesting tho to compare hardware-as-a-service or similar as well but I don't know if I see them for individual stuff.
People need to realize that when you selfhost you can choose to follow physical business constraints. If no one is in the office to turn on a computer, you're good. Also consumer hardware is so powerful (even 10 year old hardware) that can easily handle 100k monthly active users, which is barely 3k daily users, and I doubt most SMBs actually need to handle anything beyond 500 concurrent users hardware wise. So if that's the choice it comes down to writing better and more performant software, which is what is lacking nowadays.
People don't realize how good modern tooling + hardware has come. You can get by with very little if you want.
I'd bet my years salary that a good 40% of AWS customers could probably be fine with a single self hosted server using basic plug in play FOSS software on consumer hardware.
People in our industry have been selling massive lies on the need for scalability, the amount of companies that require such scalability are quite small in reality. You don't need a rocket ship to walk 2 blocks, and it often feels like this is the case in our industry.
If self hosting is "too scary" for your business, you can buy a $10 VPS but after one single year you can probably find decade old hardware that is faster than what you pay for.
I searched hetzner and honestly at just around the 500 mark (506.04) seeing it on their refurbished auction for sale, I can get around 1024 GB of ram AMD EPYC 7502 2 x 1.92 TB Datacenter SSD
In this ramflation imagine getting so much ram would cost a bank.
Like I love homelabbing too and I think that an old laptop sometimes can be enough for basic things to even more modern things but I did some calculations and colocrossing or the professional renting model or even buying new hardware model in this ramflation would probably not work.
It's sad but like, the only place it might make sense is if you can get yourself a good firewall and have an old laptop or server and will do something like this but even then I have heard it be described as not worth it by many but I think its an interesting experiment.
Also regarding the 1024 GB of ram's, holy.. I wonder how many programs need so much ram. I will still do homelabbing and things but like, y'know I am kinda hard pressed in how much we can recommend if ramflation is so much and that's when I saw someone originally writing saying 100x I really wondered how much is enough and at what scale or what others think
That’s the key. If you need one person or 3 persons doesn’t matter. The point is the salaries are fixed costs.
But all the stateful crap (like databases) gets trickier and harder the more machines you have.
But I don't know much about how it is a real world and normal 9 to 5 I have taken up jobs from system administration to reverse engineering and to even making plugins and infrastructure for minecraft I generally only work these days when people don't have any other choice and need someone who is pretty good at everything so I am completely out of the loop.
small clusters can be run with minimal knowledge which means the added cost is $0.
It would be interesting to see the scale of basecamp. I just saw right now that hetzner offers 1024 GB of ram for around 500$
Um 37signals spent around 700k$ I think on servers so if someone has this much amount of money floating around, perhaps.
Yea I looked at their numbers and they mentioned a 1300$/month for just hardware for 1.3 TB and so hetzner might still make economically more sense somehow.
I think the problem for some of these is that they go too hard on the managed services and those are good sometimes as well but like, there are cheaper managed cloud than aws etc. as well (upcloud,ovh etc.) but at the end of the day, it's good to remember that if it bothers you financially, you can migrate.
Honestly do whatever you want. Start however you want because like these things definitely interest me (which is why I am here) but I think most compute providers have really gone the path of the bottom.
The problem usually feels to me when you are worried that you might break the term of service or anything similar if you are at scale or anything, not that this stops exactly being a problem with colo but that still brings more freedom
I think if one wants freedom, they can always contact some compute providers and find what can support their use case the best while still being economical. And then choose the best option from the multitude of available options.
Also vertical scaling is a beast.
I really went into learning a lot about cloud prices recently etc. so I want to ask a question but can you tell me more about the servers that 37signals brought or any other company you know of, I can probably create a list when it makes sense and when it doesn't perhaps and the best options available in markets.
Hardware with service contracts makes sense. You can probably get the hardware even cheaper is you build supermicro servers, but then you'll spend more time on hardware support.
Dell makes a ton of sense.
How is the cost of NAT free?
> Cloud services are actually really nice and convenient if you were to ignore the eye watering cost versus DIY.
I don't doubt clouds are expensive, but in many countries it'd cost more to DIY for a proper business. Running a service isn't just running the install command. Having a team to maintain and monitor services is already expensive.
It's next to free self hosting considering even the crappiest consumer router has hardware accelerated NAT and takes a tiny amount of power. You likely already have the hardware and power since you need routing and potentially other network services
Maybe. I agree AWS is over-priced. However it shouldn't be "free".
> It's next to free self hosting considering even the crappiest consumer router
That's not the same product / service is it? We're discussing networking products and this "crappiest" consumer router wouldn't even push real world 100m of packets.
nat is free to provide because the infrastructure to have NAT is already there and there is never anything maxing out a switch cluster(most switches sit at ~1% usage since they're overspeced $1,000,000 switches), so other than host CPU time managing interrupts (which is unlikely since all network cards offload this).
sure you could argue that regional NAT might should be priced, but these companies have so much fiber between their datacenters that all of nat usage is probably a rounding error.
So it should be free? The bank already has "money". It's already there so you can take it?
That's not how it works.
Do you not get a managed service where someone upgrades it, deals with outages etc? Are those people that work 24/7 free or is it another "already there"?
And there are things that come for free when you have instrastructure this big and expansive - one-time configuration and you either monetize it or pass down the savings and since every cloud service is in agreement that profits should be maximized you end up with cloud providers which have massive datacenters at very cheap cost due to economies of scale providing it at a value far exceeding normal hosting practices due to their ability to monopolize and spend vasts amount of money onboarding businesses with false promises which errodes the infrastructure for non-cloud solutions and makes cloud providers the only choice for any business as the talent and software ends up going into maintenance mode and/or turns towards higher profitability to keep themselves afloat.
Please read it again.
> Please read it again.
There’s no need to be rude.