Parker: The new technology that allows running multiple Linux kernels without virtualization

  • Parker is ByteDance's proposal for running multiple Linux kernels without virtualization.
  • Each kernel operates on its own partition of CPU, memory, and devices.
  • The main kernel manages the hardware, while the secondary kernels operate independently.
  • The initial release does not have fault isolation between kernels.
  • Its main objective is to improve scalability in systems with a large number of cores.

ByteDance-Parker-multikernel

A few days ago, news broke that ByteDance ( creator of TikTok) surprised the free software community with the release of Parker (PARTitationed KERnel), an experimental technology that allows multiple Linux kernels to run simultaneously on a single physical machine, without relying on traditional virtualization solutions like KVM.

This advancement, proposed through the Linux kernel developers mailing list, aims to solve one of the most important challenges in modern systems: scalability on computers with large numbers of CPU cores.

Hello everyone,

Parker is a proposed feature in Linux to allow multiple Linux kernels to run.
simultaneously on a single machine, without traditional KVM virtualization. This
It is possible to partition the CPU cores, memory and devices to
partition-aware Linux kernel.

Parker, a solution without hypervisors or virtual machines

Parker's approach is based on a concept of physical hardware partitioning , dividing CPU cores, memory, and input/output devices into independent regions. Each kernel instance (called the application kernel) operates on its own allocated resources , while the boot kernel manages the initial system configuration and hardware allocation.

Unlike traditional virtualization, all Parker instances share the same kernel image , simplifying deployment and reducing execution overhead. However, each instance can have its own optimizations, configurations, and performance parameters, adapting to different workloads or use cases.

Internal functioning and resource management

Parker's core mechanism uses kexec to dynamically load new kernel instances into reserved memory regions. Before initializing these instances, the main kernel frees CPU cores, allocates physical memory using CMA (Contiguous Memory Allocator) , and detaches PCI devices that will be dedicated to the secondary kernels.

Each partition requires its own PCIe devices, such as NVMe drives or dedicated network adapters, ensuring a completely independent I/O environment. Parker also offers a kernfs-based management interface , allowing you to manage partitions and monitor the status of each running kernel.

Current limitations and future roadmap

In this first version, all kernel instances share the same isolation domain, meaning that a failure in one could affect the entire system . ByteDance acknowledges this limitation as an initial design decision, prioritizing simplicity and low overhead over complete isolation.

However, developers are already planning to leverage hardware-level isolation capabilities , such as CPU-assisted virtualization technologies, to improve fault tolerance and allow each kernel to operate in a fully autonomous and secure manner.

Although Parker shares the same goal as the Multikernel project (recently introduced by other developers in the Linux ecosystem), they differ significantly in both their architecture and technical approach . Parker adopts a shared kernel model with physically segmented resources, while Multikernel aims for a cooperative environment among multiple coordinated kernels.

ByteDance emphasizes that its implementation is completely independent and hopes the developer community will provide input and feedback to refine this technology.

It's worth noting that Parker is shaping up to be a promising experiment that could redefine how the Linux kernel scales on multi-core servers. By eliminating the traditional virtualization layer and directly leveraging physical resources, this technology paves the way for greater efficiency, lower latency , and more flexible configurations in high-performance systems.

Although still in its early stages, ByteDance's initiative demonstrates the growing interest of major technology companies in innovating within the Linux kernel, consolidating their role not only as users of the system, but also as active players in its evolution.

Finally, if you are interested in learning more, you can find the details at the following link.


Add as preferred source in Google