Home Multikernel Private Cloud Multikernel Sandbox Multikernel LiveUpdate Technology FAQ Getting Started Blog About 中文 GitHub Schedule a Demo
ARM Platform

128-Core ARM Servers: The Ideal Stage for Split-Kernel

High-core-count ARM platforms offer abundant cores and leading energy efficiency, a natural match for an architecture whose advantages grow with every added core. Multikernel partitions large many-core servers into share-nothing kernel domains, so scalability is guaranteed by the architecture itself.

More Cores, More Advantage

On a traditional single kernel, every core shares one kernel: the more cores, the heavier the global lock contention and cross-core cache-coherence traffic, and scale becomes a liability. Multikernel inverts that curve. Each kernel manages only its own subset of cores, so adding cores adds capacity without adding contention.

A 128-Core Server · Example Kernel-Domain Partition
Device kernel
16 cores
App domain A
32 cores
App domain B
32 cores
App domain C
24 cores
App domain D
24 cores

Each domain is an independent Linux kernel with dedicated cores. You decide how the machine is partitioned.

Lock Contention Doesn't Grow With Scale

Kernels share nothing: 128 cores stay as calm as 8, with no re-tuning for bigger machines.

Faults and Interference Stay in Their Domain

An overloaded or crashed domain never takes down the machine; everything else keeps running.

Full Value From Every Core

More cores means more independent domains per machine: higher density, higher utilization.

Why ARM, Specifically

The economics that made ARM the fastest-growing server architecture are the same economics that reward split-kernel.

Core Counts Lead the Industry

128 cores and beyond per socket is where ARM server silicon already lives. A single shared kernel turns that abundance into contention; a partitioned machine turns it into independent capacity.

Efficiency Is the Selling Point

ARM wins deals on performance per watt. Multikernel preserves that edge by removing the hypervisor layer, so no cycles are spent on virtualization overhead.

Upstream Linux, Not a Fork

Multikernel is built on upstream Linux with kexec-based kernel spawning and standard hotplug interfaces. Platform enablement lands upstream, where the whole ARM ecosystem benefits.

Adaptation & Partnership

A Three-Step Path to Partnership

We bring the split-kernel stack; you bring the silicon and the platforms. The work is open source end to end.

01

Platform Adaptation

Joint bring-up on your target machines: kernel boot, device kernel, and driver enablement covering network, storage, and GPU.

02

Proof of Concept

Validate performance, isolation, and operational gains on your real workloads, side by side with your current virtualization stack.

03

Upstream & Ecosystem

Push adaptation work into upstream Linux, publish joint results, and build the high-core-count ARM software ecosystem together.