We saved 30% on Kubernetes using 70% more expensive VMs | Hacker News Reader

We saved 30% on Kubernetes using 70% more expensive VMs

medium.com

3 pointsrdsubhas6 comments

I guess this epiphany meant the company did some progress on how to deploy and run their Kubernetes clusters, but the blog post reads like "we were messing our deployment strategy really badly and causing us a whole lot of problems for no good reason". I mean, calling their setup "chaos monkey" sounds like a desperate attempt to whitewash the problems they were creating for themselves.

And what was the magic ingredient? Switching to committed use nodes that can give customers up to a 70% discount? Was this supposed to be surprising?

I mean, take any intro course on the fundamentals of any cloud provider. Cost management is always one of the first topics covered. The difference between spot instance and committed pricing is rendered quite obvious.

The cloud 101 recommendation was always committed instances to cover baseline, and spin up any other instance type, perhaps spot instances, to handle peaks.

Is this what they are celebrating? That someone at the company finally paid attention to a cloud 101 course?

The article is about – saving 30% by switching from Spot to 70% more expensive VMs. It's the opposite of what you're talking about here.

> Switching to committed use nodes that can give customers up to a 70% discount?

This information is wrong. In which cloud do you see CUDs being 70% cheaper for standard machines?

Can you link any cloud provider's page that says – use "standard machines" for a higher discount than spot machines, it can be a productive conversation.

> The article is about – saving 30% by switching from Spot to 70% more expensive VMs. It's the opposite of what you're talking about here.

I don't think those values hold any meaning, specially after no concrete details are actually listed. For example, if hypothetical spot instances have a lower unit cost but your system needs to be over provisioned to account for systematic cold starts and cluster chatter, does it make any sense to say they are cheaper?

> This information is wrong.

Is it? Forward your complains to Google, then.

https://cloud.google.com/compute/docs/instances/committed-us...

Yes, that's what the article is about.

And yes, the comment is confidently wrong as said in that page.

> Yes, that's what the article is about.

Not really. Hand-waving over hypothetical unit costs means nothing. In fact, I wonder if that's why the article lists no specifics on the cost structure of the spot instance based approach, and why the cloud 101 approach is spun to be such a surprise.

> And yes, the comment is confidently wrong as said in that page.

You're literally trying to gaslight when all it takes to verify your claim is a single click. Wild.

I had asked: In which cloud do you see CUDs being 70% cheaper for standard machines?

And the article has the standard machine spec, and the page shows clearly your information is blatantly wrong.

Fitting a different content into your own pre-set wrong mental model is not worth engaging over.

Reply on news.ycombinator.com