Multikernel Linux: Multiple Linux systems without a VM on one device

Multikernel Technologies has a public version of its multikernel Linux for the first time with mklinux v7.0-mk2 published. With this, several Linux kernels can be operated on one computer at the same time and without a virtual machine.

Solution for multicore CPUs

The concept is fundamentally different from classic VMs and containers. Each kernel receives a permanently allocated portion of the available processors and RAM. This would allow, for example, a database and a GPU training process to run on the same server without each kernel accessing each other’s resources. Technically, it is not a completely independent Linux implementation, but rather a direct fork of the Linux kernel. Multikernel Technologies supplements this with numerous changes that enable the parallel operation of several kernels.

An instance acts as a host and manages the system’s hardware. When another kernel is started, the host allocates specific CPU cores, memory areas and PCI devices to it. The new kernel is then started directly on the assigned hardware using the Kexec mechanism. A hypervisor is not required. Hardware mapping is managed via device tree overlays under /sys/fs/multikernel/. According to the company, resources can even be moved between running kernel instances without having to shut them down. Modern networking and SSD hardware can also provide separate queues, giving individual kernels exclusive access to specific areas of a device.

Apps and devices separated

Another component of the architecture is the separation of applications and devices. For example, a kernel instance can only handle drivers and input/output tasks. This means that application kernels do not have to deal directly with device interrupts, which significantly improves stability. The first public release currently only supports x86_64 systems. However, the architecture-specific code is modular, so that other platforms can also be supported in the future, the manufacturer said.

The idea behind the project goes back to company founder Cong Wang, who first presented the concept on the Linux kernel mailing list in September 2025. With the now published version, the project can be used publicly for the first time and should therefore also be of interest to developers and infrastructure operators outside the company.