[numbsafari]

I have this sinking feeling that a lot of what is happening with Docker as a specific tool is going to be replaced in 4 or 5 years with unikernels.

My hope is that the orchestration/scheduling tools (mesos, kubernetes, etc.) mature in such a way that the switch from Docker to unikernels is largely transparent to most people.

[ebiester]

For all we know, that unikernel might be Docker-branded.

[nogox]

Why? unikernel is quite opinioned imo.

[numbsafari]

If you look at it from the perspective of the typical deployed application, say, with 4 or 5 VMs working together, it probably doesn't amount to much.

But if you are Google or Amazon, who have to build massive data centers to host thousands and thousands of those apps, along side much larger-scale applications, you could achieve much more significant density (and therefore reduced costs) if you were running unikernels as opposed to VMs. Perhaps passing some of that cost difference on to the customer for both competitive reasons and as an incentive for them to upgrade.

That said, even for a small-time app, consider the weight of trying to run a complicated micro-service-based system on a developer laptop. Having to orchestrate a bunch of VMs is an unmitigated disaster. Having to orchestrate a bunch of containers in one or more VMs is an improvement, but not much.

If you could instead run unikernels, there's considerably less overhead. Especially since the unikernels are typically able to run hosted inside a standard host-OS process.

Don't get me wrong, the world isn't really there. But when you consider a kubernetes cluster of docker containers that you never SSH into ... why bother with all those added layers of OS and runtime cruft?