💾 Archived View for gemini.abiscuola.com › gemlog › cloud-escapade.gmi captured on 2024-12-17 at 10:07:32. Gemini links have been rewritten to link to archived content
-=-=-=-=-=-=-
Will this be like "back to the future"?
Some days ago, I was reading some news about a company, 37signals, that boasted about saving a lot of money by leaving the cloud, and going back to "on-premise" with their infrastructure.
Previously 37signals, like many, many companies in the last decade, used to be an AWS customer and they got surprised, lately, by the amount of their cloud bills. In the end, at least by reading what their CTO said, going back to running their own hardware, helped the company save millions of dollars a year and, after going back, even more than previously projected.
In the end, what did 37signals do to save so much money? It's quite simple, perhaps: they noticed that, for their workflow, the iper-flexibility of the cloud wasn't needed and yes, part of your cloud bills goes on to pay the development, and maintenance, of all the automation and flexibility a cloud provider gives you. Given how 37signals products are built, this makes perfect sense. In essence:
Details about the final setup are not clear. Things like the hardware management and operating systems patching aren't public, but I wouldn't be surprised to read that 37signals decided to pay their colocation provider to handle that, too. It's going to be much cheaper than the cloud anyway, for their kind of setup. In the end, the main savings come from cutting a series of middle layers that, one way or another, 37signals was paying for, even if they never took advantage of those facilities.
But why is it wrong, IMHO, for a company like 37signals, to run only in a public cloud? Listing:
Given those reasons, it's obvious why they were over-paying for cloud computing resources. To add to this, they were actually getting worse performances for their buck due to the nature of the public cloud and it's shared model. Pretending to use the cloud like any other datacenter provider, is asking for troubles, when it comes to costs.
Could 37signals do something different in the cloud, to save some money there? As far as I understood, no. They were already using all the available tricks at their disposal, without revamping their applications badly to fit the cloud model.
What's surprising, are the amount of comments online claiming "exiting the cloud" would go back to hurt 37signals badly. It's like those people sees the hosting world as black or white: literally build your own datacenter, or go all in on the cloud. Such people, will defend their choices to go all-in with the cloud, with all the energies they have. However, there is a huge spectrum of possibilities in the middle.
You can, for example, build your own datacenter on a smaller scale, completing it with a small cloud deployment to cover those times where you have a burst of load. You can buy your own hardware, while renting a colocation and again, complete it with a small cloud deployment if needed. You can rent both the colocation and the hardware itself, leaving it's management to an external team or you could go all-in into the cloud, by changing your applications to be cloud-native.
The only, real, reason so many commenters defends a complete cloud deployment with blood, it's because they want, as much as possible, to shift their responsibilities to someone else. If a cloud provider has an hiccup, those IT professionals can just sit back and send their screaming bosses the direction of the cloud provider's support.
It's not a matter of "multi-region availability", or scalability, or whatever other nonsense it's thrown around. It's only, and exclusively, to get rid of as much responsibilities as possible, even if it costs more than twice as much than a proper on-premise deployment and even if a cloud, it's not the right fit for their employer products.
Is it possible to save money, on the cloud? Yes it is, but it involves going all-in with a cloud's provider managed services offerings. Managed databases, storage, serverless technologies, those are the things that, in the end, makes the cloud's elasticity worthwhile, really making you pay-per-use. The canonical "lift-and-shift" tactic, is the best way to burn a lot of money, receiving less in term of performances, while paying much more.
It's no surprise that many, many businesses are burned by their "cloud transformations", turning to things like turning off virtual machines outside business hours to save a buck (seen that myself). If you decide to go all-in in the cloud, that entails using all the possible managed services and, in the end, bad old vendor lock-in.
The smart businesses, IMHO, will either go full-managed in the cloud, or, if they have some variable workloads, they'll choose an hybrid model, where the hardware for their base load and related storage will be on-premise, complemented by a cloud deployment if the need will arise. However, given how many consultancies pitch the cloud, I also believe that the current, crazy level of cloud spending will continue, until a critical mass of businesses will be burnt during their cloud transformation journeys. I never believed in the excuse that, by going to the cloud, you could save on staffing. Only if a company had their own, complete datacenter, perhaps, but not in any other case.
This "cloud exit" is, unfortunately, an exception. The cloud is just another tool that should be used accordingly, but this is not happening right now and, for the foreseeable future, we will have fancy articles about crazy cloud spending to comment on.
