[awill]

These types of posts rarely measure operational cost. Anybody can buy infrastructure and stick it behind an API. But can you make it fault tolerant with high availability and low latency. Can you do all of that and _still_ beat AWS's costs? For the vast majority of customers the answer is no.

I used to work at a medium sized company, and they saved millions by moving to the cloud, and gained much better availability/performance. It wasn't even close, because that medium-sized company didn't have the expertise to operate the service efficiently. They just bought off the shelf stuff from VMWare etc.. Plus, DR meant paying double.

disclaimer, I now work for AWS.

[ocdtrekkie]

My availability on-prem is vastly higher than AWS, especially over the past month.

[matdehaast]

AWS/Cloud have seemed to create this fallacy that dedicated hardware/on-prem fails far more than actual reality.

[izacus]

Well, that's called "advertising and marketing" - persuading engineers to buy their products by doctoring the truth. It's actually easier with rockstar engineers - they never admit their pet approach might be wrong so AWS just needs to sell to them once ;)

[yawaworht1978]

Indeed there is an odd element to that, as when it's all said and done, AWS itself is an on-prem infrastructure too.

[awill]

when looking at these things, you always need to look over the long term. Sure, December wasn't great, but how was all of 2021 overall?

It's also not just cost/availability, but flexibility and scalability. Most high-growth startups would have not been able to scale quickly enough pre-cloud. Facebook is literally the unicorn.

[ocdtrekkie]

My datacenter has better uptime than AWS across all of 2021. Pretty sure in December alone, Amazon did worse than I've done in the past couple years combined. I have also enjoyed not being affected by various Azure or Microsoft 365-related outages throughout the year.

Scalability is a niche perk of the cloud, absolutely. I work somewhere that has a customer base that does not meaningfully scale, so it's not a concern for me. But then: Once an organization has established and has a relatively predictable scale rate, they should exit the public cloud if they can.

An established company with a stable capacity need shouldn't ever be moving to a cloud provider with a generous big tech profit margin. But that also poses a risk for the scalability of the cloud: If you don't have people overpaying for stable capacity usage, can the cloud provider afford tons of extra capacity for flexible needs?

[lixtra]

> An established company with a stable capacity need shouldn't ever be moving to a cloud provider with a generous big tech profit margin.

I’m not a fan of the cloud. But if you’re too small to have your own data center and support team, the cloud is absolutely a good option.

[izacus]

> when looking at these things, you always need to look over the long term. Sure, December wasn't great, but how was all of 2021 overall?

Overall, our 10+ year old infrastructure has hugely outpaced reliability of any of the three main clouds (and brough in more business since we got customers when the main clouds killed most of the internet).

Why do you keep behaving like reliable infrastructure was impossible without paying rent to american bigcorp?

(This doesn't mean cloud doesn't make sense for many people, but, seriously, cut the crap.)

[lifeplusplus]

No reason you can't have onsite infra but have cloud as backup for 0.009% time your stuff is down

[ddorian43]

But can they make bandwidth 1000x more expensive?

[ex_amazon_sde]

Ex Amazon here.

I've been in companies that owned their datacenters and it was much, much cheaper than using any cloud service.

Poorly managed datacenters exist but that's an organization problem. Remove the datacenter and you'll have poorly managed cloud instances and services costing millions.

[dijit]

Not sure where you get “vast majority” from, but as an anecdote I worked for two companies that operated the platform teams at 10%-15% the cost before cloud (including headcount).

However, as I’ve eluded to in other threads of this kind; people don’t like to invest in their own tech, an ideal budget would have been around 15%-20% or cloud spend, we could have solved nearly everyone’s pain with that amount of money.

DR does mean paying more, but just so we’re clear: you have to do DR in cloud too, meaning read replicas of database instances and off-cloud backups which auto restore. If your platform costs more money then this is not a good thing and isn’t just baked into the normal cost.