Multikernel Linux has released mklinux v7.0-mk2 for testing, offering an open-source alternative to VMs and containers by running multiple Linux kernels directly on bare metal.
Multikernel Linux has released mklinux v7.0-mk2, its first public test release, as an open-source project aimed at providing an alternative to conventional virtual machines and containers. It allows multiple independent Linux kernels to run simultaneously on a single bare-metal machine without a hypervisor or container service.
A host kernel divides pools of CPUs, memory, and PCI devices into instances, with each instance booting and running its own kernel natively on allocated resources. There is no emulation or hypervisor layer intercepting operations, while only explicitly provided resources are shared.
Instances are defined through device trees, with planned overlays allowing CPUs, memory, and devices to move dynamically between the resource pool and running instances without restarting. Resources can also be reclaimed when an instance is shut down.
The project is designed to provide stronger isolation than containers while avoiding some VM overhead. It can potentially assign separate kernels to applications and physical-device drivers, with virtual network and block devices enabling communication.
Initial measurements from Cong Wang, Founder and CEO, Multikernel Project, show performance gains of 1.07× to 2.5× over KVM guests, depending on the workload. Context switching was 2.5× faster, while memory bandwidth and latency were reportedly similar.
“Compared to VMs, there is no second layer of page tables and no device model,” Wang said.
The project targets AI training and inference, latency-sensitive services, and workloads requiring strong isolation. It is initially available for x86_64, based on Linux Kernel 7.0, through its public GitHub repository.
