Unikernels: The Next Stage of Linux’s Dominance [pdf] | Hacker News Reader

Unikernels: The Next Stage of Linux’s Dominance [pdf]

cs.bu.edu

94 pointsingve17 comments

This is a really exciting result. I don't have anything else to add, but this line really made me smile,

"To accomplish this, the Linux kernel has been continuously expanded to support a wide range of functionality; it is a hypervisor, a real- time OS, an SDN router, a container runtime, a BPF virtual machine, and a support layer for Emacs".

And the Linux kernel has become a subsystem of systemd ;-)

I haven't heard this before, why does emacs need a support layer in the kernel?

It's a joke. emacs is the operating system, Linux is only the kernel for emacs

Actually, its just a driver.

The real promise of unikernels is not a a "library OS", but "libraries OS". I.e. we should be combining reusable parts, not linking one huge chungus library.

I hope their initial demo really is just a straw man to get something in mainline which ends up realizing their most important longer term goal of modularizing the kernel.

Would enjoy a debate between these authors and the authors of https://www.nccgroup.trust/globalassets/our-research/us/whit... discussed at https://news.ycombinator.com/item?id=19738905

Theoretically unikernels have the potential to be secure, but the ones analyzed above did not integrate the most basic of hardening techniques. I think it's better to take that paper as "current implementations of unikernels suck" rather than "unikernels suck".

More info https://next.redhat.com/2018/11/14/ukl-a-unikernel-based-on-...

https://github.com/razaaliraza/ukl

I think it would be interesting to make a patch that turns a viable Linux kernel and glibc into just such a system. I'd want to run it off of the work of KSPP though.

This becomes really interesting once some of the proprietary software such as libcuda can be used this way.

I like the idea.

Not really, as most Unikernels don't have anything to do with Linux.

And the less of them use C, the better.

The unikernel which Ali is writing is based on Linux. That is indeed quite a big difference, but it's our belief that pulling in all the existing work done for Linux means this unikernel will ultimately be a lot more useful than the other ones.

(Disclosure: Although I'm a co-author of the paper, I haven't done much except help out Ali on a few technical things and run the meetings)

Still it feels a bit far from "Dominance".

My question might sound harsh, but did you actually read the paper ? Your comments seem a bit off regarding the actual content (not just the headline)

Yes, I did read it.

If anything what makes sense are POSIX unikernels, which as the paper mentions, are already a thing.

I don't envision a Linux monoculture, and having unikernels written in unsafe languages is contrary to the whole security idea that they bring into the table.

As proven by "unikernels" that get deployed on IoT devices, written in a mix of C and Assembly.

Reply on news.ycombinator.com