Fun article!<p>I <i>really</i> don't like that diagram. The folks doing serverless work at University of Wisconsin-Madison generally do a good job, but they really whiffed on that one.<p>I don't think there's a reasonable sense in which KVM/Qemu moves more functionality to the guest kernel than KVM/Firecracker does. Both depend on the guest kernel for VM setup (KVM), low-level virtualization operations (KVM), and doing the actual IO below their paravirtualized device drivers. On the other end, Linux containers don't have a guest kernel (unless you use some kind of library OS). So those boxes should collapse together. If you look at it one way, gVisor <i>is</i> a guest kernel, which depends on the host kernel in fundamentally similar ways to what Firecracker or Qemu do.
Unrelated to the content: I dunno what the author/site admin has done - but that was one of the fastest loading web pages I've ever seen on my phone.
To the haters— low-barrier-to-entry tinkering is the sort of thing the raspberry pi is built for. As a semi-practical example use case my raspberry pi 3b runs shairport and pihole in separate docker containers which has been a breeze to configure and has no noticeable slowness. Just being able to try new containers and services without having to worry about messing up your root environment (which would take re imaging the sd card among other things) makes it worth doing for my use. I think if you’re comparing it to a production workload you’re in the wrong ballpark.
This OTHER Firecracker is mostly obsolete now, but is still being sold. One could easily interface <i>IT</i> to a RPi as well.
<a href="https://www.x10.com/cm17a.html" rel="nofollow">https://www.x10.com/cm17a.html</a>
I could see this being useful on a Pi 4 compute (maybe), but all my work with Pis are the Zeros. There just doesn't seem like there is enough extra processing overhead to justify the simplification of deploying services instead of imaged OS.
Pi is not exactly speed daemon. I do not see much point in running containers in it. I'd rather assembler cluster and consider each Pi as a container.