Controller, Workers & the STRATUM Continuum

The controller provides the control plane and user interface. Compute workers contribute execution capacity. Storage workers contribute storage capacity. A host may combine roles where appropriate, but the roles remain distinct in the model.

Workers enroll to the controller and become part of the STRATUM Continuum. Mission Planner and the Fabrics then operate against the available distributed capacity instead of assuming every VM, image, or accelerator is local to the controller.

STRATUM officially supports mixed-architecture clusters spanning x86-64 (AMD64), Arm64 (AArch64), and IBM Z / Linux on s390x nodes in the same STRATUM Continuum. That includes enterprise customers deploying s390x alongside conventional x86-64 and Arm64 infrastructure. A single operating model connects the architectures; you do not need separate isolated management universes just because CPUs differ. This is the STRATUM way.

STRATUM makes architecture-aware scheduling decisions: guests run only on compatible worker nodes of their own CPU architecture. An x86-64 guest runs on x86-64 compute; an Arm64 guest runs on Arm64 compute; an s390x guest runs on compatible s390x compute. Live migration across CPU architectures is not supported. VM placement and same-architecture mobility remain subject to guest, VMM, device, and host compatibility.

Arm64/AArch64 is integral to STRATUM, with particular emphasis on NVIDIA Grace Hopper and NVIDIA Vera Arm-based server platforms. STRATUM has also been tested on Raspberry Pi 5 boards; Raspberry Pi and comparable single-board computers are best-effort deployments. Other UEFI-based ARMv9-A or newer server hardware is supported on a best-effort basis, as is ARMv8-A hardware in general.

Because the Continuum separates VM execution from remote resource placement, a guest running on a compatible CPU worker can be assigned a supported remote STRATUM GPU provided elsewhere in the fabric—even when that GPU host uses a different CPU architecture. This cross-architecture GPU-over-IP path does not remove the CPU architecture requirement for the guest VM itself, and still depends on compatible CUDA/provider drivers and adequate network performance.

Explore cross-site GPU-over-IP →

Continuum networking, remote storage paths, and capability-based execution placement are the foundations that let movable workloads restart on another compatible worker without rebuilding their datacenter relationships.

Read High Availability & Worker Recovery →

STRATUM Continuum can extend the same control plane and distributed execution model to ships, aircraft, satellites, remote industrial sites, field systems, and other intermittently connected edge platforms. Site-local compute and Storage Fabric keep required dependencies close to the workload while Continuum reconnects centralized operations whenever the path is available.