[zawazzi]

Historically, teams would just make their existing servers last longer. Considering most of AWS premium services have an open source alternative, guessing most people will port to containers. Built on open source has been their source of strength but is also their achilles heel. They shortened the learning curve on to their platform and off of it.

[whoisjuan]

That's a complete no-sense. When you buy from a public cloud you buy reliability, availability, security and support.

The service layer is just an add-on. Part of their premium. Even if you go full open source you still need infrastructure to run it and that's exactly what companies don't want to do and won't do anymore, and the reason why AWS and Azure are massively successful.

[zawazzi]

I think I must have miscommunicated my position. I'm not reflecting on cloud as whole, but how AWS is particularly vulnerable to a tech sector downturn. In the event of economic downturn, existing cloud customers will optimized costs (pre-cloud they would have avoided infrastructure upgrades, now they will focus on different cost structures). My belief is that generic services, such as storage, compute, ect. will priced like commodities because the market is more efficient since multiple cloud provider have the same service. Tools like terraform allow easy enough cloud configuration porting that cloud providers have little pricing power for these products. One area of cost optimization is to utilize generic cloud resources instead of AWS specific services (where I see many of my costs). Many of the services that generate positive cashflow for AWS seem to be built on top of open source which means that converting existing costly services can be converted to use generic resources with out too much implementation difficulty.

I'm only a developer so I'm sure I'm missing Operations and Business perspectives. Would love to learn more.

[owenmarshall]

Those positive cash flow services are managed. It’s cheaper to run your own Postgres instance in EC2 than it is to run RDS, until you have to pay someone to operationalize it.

You pay $.09 an hour more for an RDS m5.large than you do for an EC2 m5.large. What kind of DBA can you get for $67 a month? That’s maybe one hour of a good DBA’s time per month, tops.

Now, someone will correctly point out that you can run way more compute on a colo’d server than an EC2 instance and they’d be right. But when you realize you’re paying AWS for hypervisor patching, network automation - I’ve worked in places where a security group change in AWS was equivalent to a two week servicenow ticket with the network team - ...

The business perspective to AWS is that you bake operational costs into the product and stop paying people that generate those costs. Not good for sysadmins who aren’t willing to change, but it’s where we are. And it’s why you can’t just say “open source makes this convertible without difficulty” - somebody’s still gotta manage those services ;)

[scarface74]

The myth that Teraform makes cloud migration easy needs to die.

Have you looked at the different TF provisioners? They are all provider specific.

There are a lot of services on AWS that are specific to AWS that you would need to rearchitect your system to use. Even something as simple as an ETL process that lets you load directly from S3 into Redshift and Aurora.