I think he misses the point, or perhaps lives in another world
In my current compagny, I've a business unit with ~300 assholes (technicaly weak and afraid people), including managers etc
Question: is that enough to manager a single fucking hypervisor with a couple VMs ?
Answer: no, they are unable to perform this simple task. So for the compagny, which has too much money any way, public cloud is not in competition with "on premise". Public cloud is the only known way to have something working.
> Cloud Native, as in opposition to “lift’n’shift”, is often touted as The Real Way to take advantage of the cloud, but that’s again just more cloud marketing nonsense. It generally centers around the erroneous belief that serverless functions, and associated on-demand tooling, is going to make things cheaper.
It's not nonsense. Whether or not it is cheaper would depend.
We have gone wide with deploying all our products as SSR web apps directly on top of Azure Functions.
Our infrastructure consists of: Azure Functions, Azure SQL Server. That's it. The serverless/consumption/hyperscale tiers of each. There is no infra to manage at all. We don't even touch TLS certificates or networking anymore. There are no shared secrets req'd since functions can use managed identity auth against sql server.
Everyone on our team is able to work with these tools. Most things "just work" and are entirely managed now. For us, this is the cost savings of the cloud. The psychological cost savings are potentially more important to us than the monetary ones. Being able to focus on the customer and markets vs needy technology stacks is what gives us our competitive edge. Over the last 30 days, I have spent zero minutes worrying about database instances, VMs, operating systems, networking, firewalls, passwords, certificates, et. al. No one on our team has to think about this stuff anymore. It really feels like all of that was just a big prison and we are finally free from it.
How does focusing down an in-house technology stack add value to your product? This takes a lot of those innovation tokens and I wonder if they might be better spent.
I honestly think that you have the right attitude even though I completely agree with DHH on this topic.
Cloud has measurable value, its a fantastic option especially for prototyping or very small applications where you can handwave away all management.
As soon as the complexity creeps and you start thinking about hiring a “devops” I think thats the beginning of the end of having a good time.
Sure, some things in the cloud will always be easier and take less toil, but if spending peoples time to save $1m is possible then that money is likely better spent making your product better.
People like to talk about time as if saving a month of work will make the product better, but if you have more money in the company then that can in many cases directly tie to more headcount to solve issues.
All things are about balance, but I think we allow ourselves the comfort of not worrying about costs since we are removed enough from them. Cloud is a first class service, I thoroughly enjoy using google cloud. it makes me feel like I am magic.
But my product would not even ship if I only used google cloud… my gameserver hosting would cost €6M instead of €400,000 which is enough budget to hire many specialists to improve the game or to license expensive art, songs or work on retaining my staff.
Ironically the cloud infrastructure staff required are even more expensive than people who can manage colo servers and they insist on not even trying to save any real cost-
Another point you touched on is FaaS can scale to nothing, functionally this is the only way they can be cheaper. But if you have very high throughput and requests per second then it very quickly becomes many orders of magnitude more expensive. Scaling to zero doesn't give any savings if you have actual load because you never go to zero. (and even if you did, managed databases are another massively overpriced component which definitely does not scale)
2022, 3.2M on Cloud per year, 1M of that goes to S3 Storage.
Intend to move the $2.2M per year out of Cloud. Spend $600K on Server, amortised over 5 years. $120K per year. $60K per month on Rack, Power ( And I assume bandwidth judging from their wording ). Total = $720K + $120K = $840K.
$2.2M - $0.84M = $1.36M saving per year. $6.8M over 5 years.
One of the problem is that Compute reduction savings hasn't been passed on to customer in the recent 4-5 years.
Also worth pointing out according to recent podcast, 37signals makes tens of millions of net profits per year.
I think there is an opportunity of Cloud Exit where bare metal purchasing and handling could and should be much simpler and automatic. Instead of you owning nothing or it is simply an abstraction. One should be able to paid and has your own server shipped to you if you want it.
>Nobody is building their own data centers, outside the hyperscalers and a handful of other mega companies like Google, Microsoft, and Meta.
5 comments
[ 3.1 ms ] story [ 23.6 ms ] threadIn my current compagny, I've a business unit with ~300 assholes (technicaly weak and afraid people), including managers etc Question: is that enough to manager a single fucking hypervisor with a couple VMs ?
Answer: no, they are unable to perform this simple task. So for the compagny, which has too much money any way, public cloud is not in competition with "on premise". Public cloud is the only known way to have something working.
It's not nonsense. Whether or not it is cheaper would depend.
We have gone wide with deploying all our products as SSR web apps directly on top of Azure Functions.
Our infrastructure consists of: Azure Functions, Azure SQL Server. That's it. The serverless/consumption/hyperscale tiers of each. There is no infra to manage at all. We don't even touch TLS certificates or networking anymore. There are no shared secrets req'd since functions can use managed identity auth against sql server.
Everyone on our team is able to work with these tools. Most things "just work" and are entirely managed now. For us, this is the cost savings of the cloud. The psychological cost savings are potentially more important to us than the monetary ones. Being able to focus on the customer and markets vs needy technology stacks is what gives us our competitive edge. Over the last 30 days, I have spent zero minutes worrying about database instances, VMs, operating systems, networking, firewalls, passwords, certificates, et. al. No one on our team has to think about this stuff anymore. It really feels like all of that was just a big prison and we are finally free from it.
How does focusing down an in-house technology stack add value to your product? This takes a lot of those innovation tokens and I wonder if they might be better spent.
Cloud has measurable value, its a fantastic option especially for prototyping or very small applications where you can handwave away all management.
As soon as the complexity creeps and you start thinking about hiring a “devops” I think thats the beginning of the end of having a good time.
Sure, some things in the cloud will always be easier and take less toil, but if spending peoples time to save $1m is possible then that money is likely better spent making your product better.
People like to talk about time as if saving a month of work will make the product better, but if you have more money in the company then that can in many cases directly tie to more headcount to solve issues.
All things are about balance, but I think we allow ourselves the comfort of not worrying about costs since we are removed enough from them. Cloud is a first class service, I thoroughly enjoy using google cloud. it makes me feel like I am magic.
But my product would not even ship if I only used google cloud… my gameserver hosting would cost €6M instead of €400,000 which is enough budget to hire many specialists to improve the game or to license expensive art, songs or work on retaining my staff.
Ironically the cloud infrastructure staff required are even more expensive than people who can manage colo servers and they insist on not even trying to save any real cost-
Another point you touched on is FaaS can scale to nothing, functionally this is the only way they can be cheaper. But if you have very high throughput and requests per second then it very quickly becomes many orders of magnitude more expensive. Scaling to zero doesn't give any savings if you have actual load because you never go to zero. (and even if you did, managed databases are another massively overpriced component which definitely does not scale)
Intend to move the $2.2M per year out of Cloud. Spend $600K on Server, amortised over 5 years. $120K per year. $60K per month on Rack, Power ( And I assume bandwidth judging from their wording ). Total = $720K + $120K = $840K.
$2.2M - $0.84M = $1.36M saving per year. $6.8M over 5 years.
One of the problem is that Compute reduction savings hasn't been passed on to customer in the recent 4-5 years.
Also worth pointing out according to recent podcast, 37signals makes tens of millions of net profits per year.
I think there is an opportunity of Cloud Exit where bare metal purchasing and handling could and should be much simpler and automatic. Instead of you owning nothing or it is simply an abstraction. One should be able to paid and has your own server shipped to you if you want it.
>Nobody is building their own data centers, outside the hyperscalers and a handful of other mega companies like Google, Microsoft, and Meta.
Google, Microsoft and Meta ARE hyperscalers.