An absolute number without even mentioning the period is clickbaity. I'd like to know a percentage.

Our startup suffers from our AWS bill, but it's well below 10k a month. So we would be glad to reduce it by 3k or 5k.

Considered bare metal at all? Also have a startup - we estimated our cloud bill would have been over $4k/mo+ but on bare metal we are running at about $200/mo

Sure, we have. But reliable people to run your infra are basically impossible to get at reasonable cost. You'd pay the one sysadmin/devop the same we pay AWS. If she/he leaves you are in deep trouble.

And you still have not bought the HW...

Wouldn't it be possible to pay a company to setup and manage the bare metal infrastructure for you so as to reduce costs (they have multiple customers) and increase redundancy (the consulting company has more than one employee)?

Well, AWS is our "consulting company" now. Maybe they are not the absolutely cheapest one. But they are easy to work with if you are small.

Yes, you need to think that you don't go into the worst cost traps. But no form of subcontracting/purchasing frees you from making the right choices.

Cloud watch charges are the most enraging thing. What are they even charging for, storing and filtering some text files? I have seen services spending 30% of their costs on cloud watch. Absolute madness.

Also Not Awesome - AWS Config rule evaluations

Maybe I should write up how we saved > 500k/yr on our AWS bills if folks find that stuff interesting.

There are known techniques, specifically reserved instances and spot instances, but your $500k savings isn't the same as someone else's $500k savings, so if you feel so inclined, you should definitely do it!

Thank you for the kind words. At the very least it would give me space to explore my experience and document it.

Go to the cloud, it's cheap. And when it's not you just turn off parts of your system.

I am somewhat suspicious that that's not really the solution.

Cloud is not cheap.

You trade "waste", and high up-front cost, for a significantly hightened sustained cost.

You trade setting up systems and proper prior planning for understanding proprietary interfaces and understanding cost systems.

It's too far to say it's a scam: it's a trade-off.

There are much cheaper on-prem solutions and there are cheaper cloud systems.

The difference is that the cloud ones TCO is only lower at either absurdly small volumes or absurdly spiky loads (IE; a website that only operates at full load for 3hrs a day and is otherwise under 1 request per hour)

The savings is with not having to deal with HR to manage the people hired to keep your local systems up and running.

The tradeoff is that if you have a sudden hit event you may simply not have the raw horsepower to handle the influx, but you also can't be hit with surprise $100k+ bills for a configuration mistake.

That's sales speak.

Kind of interesting watching it add up. Any optimization on its own would be pointless.

I struggle with this. The only cost savings projects I've had success pushing forward are when it's fairly large. But I wouldn't be surprised if most waste is kind of like a death by a thousand cuts where it's to expensive to address them individually.

For some of the these, if you use a engineers pay of 100/hr, if you want to save 2000 a year you need to ship the whole thing in 20 hours, just half of a week of work. It's almost not worth talking about using the hourly rate.

I think you're looking at the problem with some incorrect assumptions. First, Amazon wants us to believe that people who can run things in AWS are just too dumb to be admins. I think they want this to be the case. After all, programmers who know nothing about the underlying systems their software runs on are the kind who'll turn up services and leave them on. They'll build something that has the kind of fragility that makes others afraid to turn things off to save money. Each new thing will add more and more to the monthly bill.

A good systems administrator works to regulate this tendency. She/he asks what these processes are, whether all this used storage will be needed indefinitely, what the dependencies are between different services. A good sysadmin balances the laziness of programmers who don't want to think about or care about systems long term.

Marketing for moving everything to the cloud also has tons of kickbacks that encourage decisionmakers to want to fire sysadmins and pay Amazon so they can get free stuff. Is it completely nuts that someone will have their company pay hundreds of thousands of dollars a year to Amazon because they get a few thousand dollars worth of perks? Sure, it's nuts, but people are greedy and selfish, so this happens all the time.

The point is that from a purely technical and mathematical point of view, cloud loses every time except perhaps for the most ephemeral of uses.

  I think you're looking at the problem with some incorrect assumptions. First, Amazon wants us to believe that people who can run things in AWS are just too dumb to be admins.

On the other hand, most people I meet at work are too dumb to be admins. Sad, but true :/

To the author: can you elaborate on situations where the sales tax is recoverable?