Forgot your password?
typodupeerror
Linux

Multikernel Linux Tree Released: Runs Multiple Kernels On Bare Metal Without a Hypervisor (itsfoss.com) 36

"Multikernel Technologies has put out mklinux v7.0-mk2, the first public release of the multikernel Linux tree it's been building since last year," reports the blog It's FOSS: The basic premise of the project is that one physical server can run several independent Linux kernels side by side, each with its own dedicated CPUs and memory. So, the same box could run a database on one kernel and a GPU training job on another without either kernel touching the other's resources... One kernel runs as the host, controlling the machine's CPUs, memory, and PCI devices. When it spins up a new kernel instance, that instance gets its own dedicated slice of hardware and boots straight onto it using Linux's kexec mechanism, running natively rather than under a hypervisor.

The host tracks all of this through device tree overlays under /sys/fs/multikernel/, and it can shift hardware between running instances while everything stays up.

That's a different setup from both VMs and containers. A VM still puts a hypervisor and a host kernel between the app and the hardware. A container skips the hypervisor, but every tenant on the machine shares one kernel... This release only supports x86_64 hardware, though Multikernel Technologies says the architecture-specific code is already modular enough for other ports to follow later.

The article notes they're selling multiple products based on the kernel tree, including one for running AI agents in their own kernel with direct GPU access and one that promises to patch a running kernel without taking the system down.

Multikernel Linux Tree Released: Runs Multiple Kernels On Bare Metal Without a Hypervisor

Comments Filter:
  • Can it power off CPUs and other devices allowing for hot-swap? I can imagine the next generation of multi-cpu/conntroller motherboards with power LEDs for each socket.

    • by tlhIngan ( 30335 ) <slashdot@@@worf...net> on Monday August 31, 2026 @12:30AM (#66315178)

      I believe Linux already supports CPU hot-swap if the hardware supports it. It was needed by IBM so their POWER mainframes could literally do just that.

      Unfortunately I don't think there's many x86-based servers that support CPU hot swapping.

      • Physical ones are definitely rare beasts. Hypervisors make having support for it a lot more generally useful, though. You can still run into weird edge case limitations(certain PCIe devices, under some circumstances, really don't like being detached from the host and popped onto a VM); but the hypervisor will definitely be able to tell guests stories of dynamic reconfiguration that the hardware isn't electrically or mechanically capable of supporting; and being able to add or remove CPUs or RAM without rebo
    • by thegarbz ( 1787294 ) on Monday August 31, 2026 @12:54AM (#66315188)

      Can it power off CPUs and other devices allowing for hot-swap?

      Linux introduced CPU Hotplugging in 2.6.16 almost exactly 20 years ago.

      • Can you save me the time googling this and point me to a hardware manufacturer who sold systems that supported this?

        • by Filgy ( 2588 )
          I was going to guess that IBM or any blade server provider does but the AI tells me that they all require swapping of the entire compute node/blade. So it looks like Linux has supported it since forever but there are no offerings that take advantage of it if you want to hotswap ONLY the CPU.
          • I do recall HP (before the trailing 'e') having some... specifically their top end Itanium systems had hot swap CPU in the hardware. It was developed to compete with the big iron mainframes so that sort of functionality was needed. I remember standing up one of these systems, and because I never actually used that functionality I don't know how it worked or how well it worked but it was definitely a feature I remember the HP touting when we purchased them.

            For all its warts, if you ran native code on them th

            • Where Itanic sunk is because Intel were heavily pushing the x86 compatibility of the processor

              That and the price per op, which was alone enough to make it a flop.

        • Mostly big iron like mainframes.

        • IBM Power systems since POWER4 have been able to attach and detach CPUs, Memory, and I/O (virtual or physical) to a running LPAR.
        • by keltor ( 99721 ) *
          Hot swap CPUs are always on cards.
        • by _merlin ( 160982 )

          Various SPARC-based Sun Enterprise Server products supported hot-swapping CPUs under Solaris. I think some of the POWER-based IBM System p servers support hot-swapping CPUs under Linux. Maybe some of the Unisys ES7000 support it?

  • Most modern servers have more than one core.

    Not that hard to treat each core as a separate machine, all using another core as something like an internal router to access memory, graphic cards, keyboards, etc.

    The hardware changes would be minimal. Basically take 4 motherboards, put it one metal box and put most of the real memory, etc. on the 'host' motherboard that allocates between the other 3.

    There would be real speed penalties to doing this, but totally possible just using hardware. Duplicating it with

    • Doing it in software means you can move CPUs from one machine to another based on the time of day. Your CI infrastructure can lose some resources from its UI at night and make another server for nightly builds when most of your users are away, without either having a VM layer in between or needing the builds to happen on a kernel configured with the features needed by the UI.

      • by dfghjk ( 711126 )

        "Doing it in software means you can move CPUs from one machine to another based on the time of day."

        Which you could do with a single OS much more effectively.

        "Your CI infrastructure can lose some resources from its UI at night and make another server for nightly builds when most of your users are away, without either having a VM layer in between or needing the builds to happen on a kernel configured with the features needed by the UI."

        But not without restarting and with the limitation that different "system

  • by Anonymous Coward on Monday August 31, 2026 @12:12AM (#66315158)

    IBM mainframes have LPARS - logical partitions - to divide up the hardware. This sounds like almost the same thing, but for PCs.

    https://www.ibm.com/docs/en/zos-basic-skills?topic=design-mainframe-hardware-logical-partitions-lpars

  • by nester ( 14407 ) on Monday August 31, 2026 @12:25AM (#66315172)

    As someone who worked a bit on mklinux (after Apple abandoned it), I wonder if they googled before choosing a name. The mklinux.org website is actually still up, amazingly.

  • Without the 300 baud jtag interface.

  • This is interesting, but I'm trying to figure out what the real-world applications of this might be. What the purpose or point of this?

  • ...and a former Mainframe op, this means very little to me unless / until it has a decent TUI and/or GUI interface to report on, configure, and manage things.

    Give me an ISO to boot and an interface as described above, and I'll dedicate some time to testing it.

  • This sounds like a kernel that's also a hypervisor. Corrections are welcome!

Counting in binary is just like counting in decimal -- if you are all thumbs. -- Glaser and Way

Working...