It seems like multikernel technologies’s solution of kernel on dedicated cores but shareing hardware and kerbel IPC adds more novel use cases (like live kernel hand offs!) But this is probally better in isolation caps (closer to trad vms).
man the scale of proc isolation from unikernal to kernel mods to ebpf to threads to children to forks to seperate procs to wasm to cgroups to containers to lxd to multi kenerl with ipc and now parker and finally full on virtualized hardware.
All of that running in combinations on the same CPU. Itll be really cool to see what kind of abstractions we will make to make it easier to compose all that into patterns that are useful.
- Man, it’s so annoying to have to use multiple machines because one isn’t beefy enough!
- We can now shove the hardware of 10 machines into one, making it the beefiest!
- Man, it’s so annoying to have to use such a beefy machine!
- We can now divide a beefy machine into 10 mediocre ones, just like you always wanted!
Make it make sense to me.
Not sure if you’re being sarcastic or not, but the point of this is to reduce overhead associated with virtualization (aka VMs). Few workloads are able to take advantage of the massive compute resources that a single beefy machine has, so partitioning it is the most efficient use of resources, especially in data centers where maximizing efficiency is important.
Sometimes you have different use cases. One beefy machine is likely more energy efficient. Many small machines allows reuse of old hardware.
I wish we’d get a proper microkernel, instead. If we’re going to accept some performance degradation, let’s get some benefit from it.
Again: I have high hopes for Redox.
Why keep ten machines when you can do the same job with one?
