↓ Skip to main content
Dolysis Consulting
  1. Posts/

The Cloud Works For Itself, Not You

509 words·3 mins
opinion

Farah Schüller of 37signals released a post today on their experience and savings with migrating from Docker Hub (and originally AWS ECS1).

You should read Farah’s post; it’s succinct, it’s pertinent, and it’s fun.

Back? Great, here’s the points we’ll be talking about below:

  • Switching to a self-hosted Harbor image registry reduced their yearly infrastructure bill by ~$5000, improved availability, and unlocked advanced retention rules not available in ECS leading to a ~80% (7.5 TB) reduction in storage wastage
  • Switching simplified their infrastructure, lowering the maintain burden for operations.

The increase in savings is great—just think about what your business could do with an extra $5k—but it’s not what we are going to focus on.

The Cloud Is Not Simple
#

Does anyone else remember being sold that the Cloud would simplify and streamline your infrastructure? There was also the additional, implied statement that you can fire your IT division if you move to their platform.

Well, here we are 15 years later, and between the…

  1. Billing (AWS CostExplorer is so good, you have to hire two external firms to parse it for you)
  2. Visibility (CloudWatch has substantially worse features than Zabbix)
  3. Networking2
  4. Security (every minute spent debugging confusing SG rules makes the business $0. *Zero.)

We can see that, well, maybe it’s not more simple but at least we don’t have to pay as many people for… oh? Wait, it basically created a new industry of Cloud engineers, leading a general increase in engineer hours spent on operations work?

Okay, so we have to pay more people to use more expensive, more complicated software; giving AWS a ~40% profit margin3… but, I mean, the sales slide deck was really nice.

This is not sustainable for normal businesses. What we need is stable pricing and boring4 software. Most businesses in the Cloud aren’t using fancy (and extremely expensive!) services like CloudWAN or multi-writer filesystems. We’re deploying Rails or Java—or Python, or Node.js, the tech isn’t important—apps to help our customers.

IT infrastructure should fall into the same category as a good hammer: it should be a value multiplier, that we buy once and use until we need to replace it, and is operated by a tradesman that knows how to use hammers well.

Your IT infrastructure should for work you, not your provider’s shareholders.

  1. https://dev.37signals.com/bringing-our-apps-back-home/ ↩︎

  2. Oh, so you want to connect two networks together? Well, don’t use subnet peering or you’re going to have a ~10k bill for ingress/egress traffic (double charging, baby)! PrivateLink? It’s the recommended solution for point-to-point. That’ll be $22.11 per 3az deployment even if you send 0 packets. Hope you don’t have any many-to-many traffic patterns, or the cost graph goes exponential. Well, how about TransferGateway? Soft-deprecated in favor of CloudWAN, which requires you to use an entirely new IAM service because it was too complex to write the rules for it using plain old IAM. Surely, this is simpler than… running two VLANs over a Juniper switch, which has continued to function the exact same way for 25 years now. /endrant ↩︎

  3. https://hyperframeresearch.com/2025/05/02/aws-profits-surge-powering-amazons-future-growth ↩︎

  4. https://mcfunley.com/choose-boring-technology ↩︎