The Cloud is Just Another Sun [video]
fosdem.org
fosdem.org
Furthermore, S3's replication is "eventually consistent", something that you wouldn't accept in a filesystem-like use.
Source: the internets, and I was at AWS for 6 years (2008-2014).
I'd still call this a huge win for both open source and free software. But obviously its a much bigger win for open source than for free software.
Linux didn't beat Sun because it was better, cheaper, or had more ideologically pure licensing.
Linux dominated because it was accessible. Anybody could download it, try it, it ran on most things. Barrier to entry was essentially zero. Licensing helps, but it's being trivial to access that is the killer outcome.
And the reason behind the rise of cloud isn't because it's cheaper or better, it's that it's more accessible. Anybody can get out their credit card and try it, use it. Barrier to entry is essentially zero, and that's how to suck people in.
It’s easy to see this turning in a more proprietary direction later in the game, with GKE leaving Kube itself behind, but for now it seems like a strong contender in the fight against “everyone just codes against their preferred cloud’s API.” Commoditize your complement, etc.
Every cloud platform needs its own management layer to make the hardware infrastructure useful, and it's at this management layer where lock-in happens. Google made this layer a commodity with k8s, but took it a step further by making the trend toward open infectious and one-way.
Google's gamble was that if providers have to compete directly on a level playing field, Google will win because they're better at it. It's a bold move, and we'll eventually see if they were right. In the mean time k8s is still at work leveling the field.
Here's the key that makes k8s a one-way de-proprietarization mechanism: it's not just infrastructure for building systems, it's infrastructure for building infrastructure, and it's better than anything you could come up with otherwise, and the infrastructure it builds is natively open and extensible. K8s is trivial to extend, but it's REALLY HARD to extend it to become a proprietary system, while still keeping it useful.
This is true for gke as much as anything else. The value-add for gke is that someone is managing it who knows wtf they're doing. The value add isn't (and can't be) any Google-specific magic sauce they add to k8s, because k8s is only really useful if the customer themselves can extend it using known interfaces. And you can't make gke meaningfully proprietary without breaking that extensibility. Sure they can have gke-specific components, just like how Amazon can have aws-specific controllers if they need to. But those have to be extensions to an open base. If such things could result in lock-in, then they would also result in a lock-out of common tools and extensions, which would kill the platform dead.
The luxury of having that problem is pretty nice too. The benefits of working off or with these tools is you can get something up and running way faster than owning the whole stack yourself. If the services are priced as a utility then theres no reason not to use them. If your business is running on margins that require better or different systems then thats predictable and you can just design for that.
A company offering great prices now might chose to increase them later on when they have achieved market leader position and everyone is locked into their proprietary APIs. Don't assume that their offers will always remain as sweet as the bait candy that you taste right now.