Posted to the Linux kernel mailing list one month ago were patches for a multi-kernel architecture design to allow multiple independent kernel instances to co-exist on the same single physical machine. This could let some CPU cores be running real-time “RT” kernels or other non-traditional uses between CPU cores. It wasn’t clear how far the multi-kernel patches would get especially with some initial negative views toward it and Bytedance separately proposing “Parker” for multi-kernel usage just days later. In any event, today a second version of the multi-kernel Linux patches were posted.
It’s still not clear if there is enough traction and interest from the key Linux kernel developers to get the multi-kernel architecture eventually into the mainline Linux kernel. The reported benefits around better resource utilization, better performance than containers, improved fault isolation, and other stated benefits still need to be better showcased.
In any event Cong Wang with Multikernel Technologies Inc today posted the v2 patches for getting out the newest code. This new code introduces generic multi-kernel physical memory allocations and per-instance virtual memory allocations, a kernfs interface for managing the kernel instances, Device Tree and Kernel Handover (KHO) framework for resource management and sharing, messaging via IPI, and reorganized/improved patches.
They have also now posted a YouTube video for demonstrating the multi-kernel architecture:
The v2 patches are still out under a Request For Comments flag. Those wanting to learn more about this proposed multi-kernel architecture for Linux can see the new patch series.
