Large shops seem to love complexity and mistake it for flexibility while constantly complaining about how complicated things are to integrate. It puzzles me how the culture seems to keep adopting stuff so complicated that only those with an army of consultants and massive resources could do the same thing while most start-ups look at the same problems, find a solution that gets the problem solved sufficiently and then build things that actually solve real problems instead of self-inflicted ones like dealing with the complexities of cross-account permission structures via AWS IAM and S3 bucket policies interacting with SSO.

I’m not entirely convinced that the F500 would adopt GCP if they had the support and hand-holding structure that they have historically demanded. Myself, I’d rather find a solution that doesn’t need such a team to support me in the first place. Sure, there’s all sorts of issues with partners’ service offerings and the high cost of transit between multiple cloud providers but perhaps you can save the costs of support and put it toward network instead (won’t solve the transit costs for your partners though).

The complexity of AWS is exactly why it is so successful.

It is being pushed by both outside consultants and employees who become the experts within the enterprise. GCP's ease of use is a disadvantage in the big picture.

This is a pragmatic and astute observation. This scenario has played out several times in the past.

> This scenario has played out several times in the past.

What examples come to mind?

VMware

Enterprises tend to buy full IT offerings including apps and services, not just IaaS. Having a big partner network, which VMware does, is a competitive advantage. Amazon and Azure likewise have large partner ecosystems.

Disclaimer: I work for VMware.

Oracle.

That is what makes Oracles cloud play such a mystery. They had a huge asset in a loyal corps of Oracle DBAs embedded in every large organisation on the planet. Now they are saying “fire those guys and we’ll host your databases”. But maybe those guys will say “nah, Postgres is actually all we need, see ya”.

Oracle had no choice. They had to disrupt their own business or someone else would do it to them.

Right, but they could have upskilled/cross-trained the DBAs to be cloud guys, offered training, certificates, conferences, yadda yadda. Instead they chose to stab their strongest advocates in the back!

A smart executive (ha!) would announce that next year’s bonus pool would be funded by savings on Oracle licences then sit back and let nature take its course...

Bloomberg terminals

Netware

GCP is only easy to use because it has no features compared to AWS.

GCP is not less complex. It just appears (but is not actually) better organized and documented. You still need an operations army to fill in feature gaps or go beyond anything that is not default, which are the much bigger problems. At least on AWS you can hack it to do what you need, on GCP some things you just cannot do.

What's an example of an intractable problem you ran into on GCP?

Maintenance windows that don't cause downtime? This was even discussed here:

https://news.ycombinator.com/item?id=16871931

Read the last description: "amateur and unprofessional." The same thing happens for Kubernetes Engine. Does Google internally run their Cadillac infra like this? Of course not. The conclusion is either they don't care about their external cloud or that team can't ship basic features.

I'm not sure I follow you here: for GCE, there are no "maintenance windows", the vast majority of maintenance is carried out via live migration and the user doesn't even notice.

Yes, on rare occasion there's an unplanned failure (host error) and the VM has to be restarted elsewhere, but this is the only option for maintenance at AWS.

Mandatory disclaimer: I work at GCP.

As asey correctly points out, the link I posted talks about the Cloud SQL maintenance window, not about Compute Engine. That there is unavoidable downtime even with HA has been a problem known for seven months at this point. A Cloud SQL TL even sought feedback here with people telling that person this is a "dealbreaker" and another saying it doesn't even make sense to have built HA for a low probability event (zone outage) before dealing with this much higher probability event:

https://news.ycombinator.com/threads?id=flamingcow

This is pretty strong customer feedback. What happened with this feedback, anything? The bigger picture here about GCP, preceded by Google's long reputation, is that customer feedback means nothing. Googlers are probably still building what they feel like building, and that is why AWS continues to trot ahead despite being less polished. They ship things that work and they know what their customers want.

One example fitting GP description would be Cloud SQL maintenance windows. Unavoidable downtime as far as I know (even with HA)

FWIW if you're depending on something to be up in a single availability zone, you're already doing cloud wrong.