4 ms·

Cloud Costs Every Programmer Should Know

27 points by SerCe 3y ago

DarthNebo 3y ago
Feels like there should be two branches of system design for unit profitable architecture for paying users vs VC backed architectures to support non-paying users.
withinboredom 3y ago
This is why you don’t run in the cloud. I’m paying (amortized) 300 bucks a month for a physical k8s cluster, multi-region, hundreds of physical cores, terabytes of disk space, and oodles of RAM (64gb per machine). The same cluster would cost over ten grand a month in the cloud, for what? Scaling? I’m actually comfortably overprovisioned. People underestimate how many requests a (real) server can handle. For the vast majority of businesses, the cloud is overkill and expensive. Sure, it means that you have to drive somewhere and replace some hardware instead of pushing a button, but most places offer “remote hands” to do that for you.
- Our_Benefactors 3y ago
  There’s a lot more overhead than just replacing the hardware. You have to manage the rest of your software stack as well; no vendor will update log4j vulns or let you know when your k8s needs an upgrade. For some users the cost of cloud is worth it for these kinds of reasons.
  - withinboredom 3y ago
    Because the cloud automatically upgrading your cluster is better? I’ve seen that take out an entire cluster more than once.
    - Our_Benefactors 3y ago
      Maybe it automatically upgrades if you’ve configured it that way - I wouldn’t recommend that either! But using a managed solution will likely give you more options for resolution, and more awareness of problems as they occur - for a self managed cluster you’re totally on your own in terms of all support and maintenance.
      - withinboredom 3y ago
        How so? I just read the release notes on the helm chart and underlying process and bob's your uncle. I file issues when I find them and wait for them to be fixed before upgrading the main cluster (most recent one was an issue with ingress-nginx, but I've also found issues in Cilium, git, Linux networking stack, etc.) Things break, you diagnose the issue and roll back. There's nothing special about it, and it is likely the exact same thing you do when releasing any of your own software -- so there's nothing new here, at all. What magical options do you have with a managed solution that I don't have?
- LorenzoGood 3y ago
  Where?

27 points by SerCe 3y ago

Feels like there should be two branches of system design for unit profitable architecture for paying users vs VC backed architectures to support non-paying users.

withinboredom 3y ago

This is why you don’t run in the cloud. I’m paying (amortized) 300 bucks a month for a physical k8s cluster, multi-region, hundreds of physical cores, terabytes of disk space, and oodles of RAM (64gb per machine). The same cluster would cost over ten grand a month in the cloud, for what? Scaling? I’m actually comfortably overprovisioned. People underestimate how many requests a (real) server can handle. For the vast majority of businesses, the cloud is overkill and expensive. Sure, it means that you have to drive somewhere and replace some hardware instead of pushing a button, but most places offer “remote hands” to do that for you.

Our_Benefactors 3y ago
There’s a lot more overhead than just replacing the hardware. You have to manage the rest of your software stack as well; no vendor will update log4j vulns or let you know when your k8s needs an upgrade. For some users the cost of cloud is worth it for these kinds of reasons.
- withinboredom 3y ago
  Because the cloud automatically upgrading your cluster is better? I’ve seen that take out an entire cluster more than once.
  - Our_Benefactors 3y ago
    Maybe it automatically upgrades if you’ve configured it that way - I wouldn’t recommend that either! But using a managed solution will likely give you more options for resolution, and more awareness of problems as they occur - for a self managed cluster you’re totally on your own in terms of all support and maintenance.
    - withinboredom 3y ago
      How so? I just read the release notes on the helm chart and underlying process and bob's your uncle. I file issues when I find them and wait for them to be fixed before upgrading the main cluster (most recent one was an issue with ingress-nginx, but I've also found issues in Cilium, git, Linux networking stack, etc.) Things break, you diagnose the issue and roll back. There's nothing special about it, and it is likely the exact same thing you do when releasing any of your own software -- so there's nothing new here, at all. What magical options do you have with a managed solution that I don't have?
LorenzoGood 3y ago
Where?