Oxide Computer Company, a developer of rack-scale hardware, has launched 'Virtual OXidE Labs' (Voxel) to address the logistical challenges of testing and developing large-scale infrastructure. Because completed racks are often deployed immediately to customer data centers, the company faces a shortage of available hardware for internal testing. Voxel provides a high-fidelity software emulation stack built on illumos in Rust, designed to replicate diverse customer networking topologies and run the company's actual control plane software without requiring physical rack space.
The image serves as a thematic illustration for the 'Virtual OXidE Labs' (Voxel) software emulation stack.
The Voxel environment simulates several core components of the Oxide ecosystem. It provides sled virtualization via Propolis, supports BGP networking using Debian and FRR, and emulates P4-powered switches through Oxide’s own SoftNPU. Additionally, it simulates microcontrollers for the Service Processor (SP) and Root of Trust (RoT). This allows the company to maintain a 'Prime Directive'—testing stock Oxide software without modifications—while running the entire stack on relatively modest hardware. For example, the author of the report notes that a full rack configuration with six sleds can be spun up in approximately 10 minutes on a single mini-PC equipped with a 32-thread AMD Ryzen 9 processor and 96GB of RAM.
To ensure the emulation remains accurate to production environments, Oxide uses a snapshot-based workflow. Developers can create a 'sled' virtual machine, install the Omicron control plane, and take a ZFS snapshot before configuration. These snapshots can be reused to quickly instantiate new racks. The system also supports different installation methods, such as cloning the Omicron repository directly from GitHub or pulling pre-packaged releases via The Update Framework (TUF). By using these virtualized environments, Oxide can automate testing and simulate complex upgrade scenarios, which the company identifies as a critical requirement for finding regressions and verifying new functionality in rack-scale systems.