Public clouds provide on-demand computing resources over the Internet. The largest are called hyperscalers.
Almost all hyperscalers, at the notable exception of Microsoft Azure, depend on open-source software to support their platform . Put simply, without open-source software, they wouldn't exist, at least not in their current form.
These providers also add their own custom software to the core engine that propels their platform . Unfortunately, this custom software remains for internal use only.
Isn't there an equivalent to these custom building blocks available for everyone to reuse?
Rust-vmm (or Rust-Virtual Machine Monitor) is an ongoing effort among software and hardware companies, including some hyperscalers, to share more of their codebase. Rust-vmm provides a platform to share reusable virtualization-related code by means of Rust-crates .
As of 2021, this project offers the closest open-source equivalent to the aforementioned custom software used by hyperscalers.
At least three key projects using Linux and KVM are also taking advantage of Rust-vmm :
crosvm (2010 --)
firecracker (2018 --)
Cloud Hypervisor (2019 --)
|Support for non-Unix guests||No||No||Yes|
Until recently, any attempt to create a local-first, free and open-source operating system that could run atop affordable, virtualization-friendly hardware using basic building blocks similar to those used by major public clouds would rightfully be met with skepticism.
Thanks to the rust-vmm umbrella project, assembling such an operating system is now becoming a possibility.
Phyllome OS intends to tap into some modern software- and hardware-related innovations used in the cloud and make them available to a wider audience locally: to bring some of the cloud back home, so to speak.
Phyllome OS draws inspiration from numerous other projects, including desktop-oriented systems such as Qubes OS, Tails, and Fedora Silverblue, as well as others specialized in running container workloads, such as Fedora CoreOS and RancherOS.
When it comes to virtualization-friendly, open-source, desktop-oriented operating systems, two projects stand out: Qubes OS and Spectrum. How do they compare to Phyllome OS?
Like Phyllome OS, Qubes OS is based on Fedora but relies on Xen, the other popular open-source hypervisor for Linux.
Xen strongly isolates components of the hardware stack, including the USB and network controllers. By design, it works in parallel rather than alongside Linux, as KVM does. KVM’s more tight integration with the Linux Kernel can be considered an advantage or a disadvantage.
Out of security concerns, Qubes OS does not yet support 3D-accelerated virtual machines, even though its parent project Xen does support this functionality. Phyllome OS intends to support 3D acceleration inside virtual machines, even if it means increasing the attack surface.
Just as with Qubes OS, Spectrum’s main focus is secure computing. Spectrum uses Nix, a declarative packet manager. It is built atop crosvm and thus doesn’t rely on QEMU, largely reducing the attack surface. Through a re-implementation of the virtio-wayland device, which is used in Chrome OS to securely run Linux apps alongside the main OS, Spectrum will eventually allow its guests’ virtual machines to have a GPU capable of efficiently accelerating 3D applications.
By design, Spectrum won't support operating systems that don't rely on the Wayland protocol.
|Qubes OS||Spectrum||Phyllome OS 1.0|
|Virtual chipset||i440fx? / Q35?||?||virt|
|Non-Linux guests support||Yes||No||Yes|
|Based on||Fedora||Chromium OS?||Fedora CoreOS|
|Desktop Environment||Xfce||Aura?||GNOME Shell/Headless|
|OS as the center of the UX||Yes||Yes||No|
|Portability of VMs||No||No||Yes|
From a design perspective, Qubes OS and Spectrum are end-to-end operating systems, whereas Phyllome OS is only a wrapper around the user’s preferred operating system. Thanks to nested-virtualization, it could even be used to host those operating systems, but in this configuration, the attack surface would be significantly increased, and the performance would take a significant hit, especially for nested guests.
In Phyllome OS, the main computing activity will happen inside the user’s virtual machine. In QubesOS, Dom0 (“domain zero”) is at the center of the user’s experience.
In summary, despite some shared characteristics, Phyllome OS is not meant to be a replacement for Qubes OS or Spectrum, but could become a test bed for these operating systems.
Since 2017, Xen, upon which Qubes OS relies, is also exploring the possibility to avoid using QEMU for guests using hardware-assisted virtualization. See the diagram on the “Guest Types” section:“Xen Project Software Official Overview.”. ↩︎ ↩︎
Open-source or free software are software that are freely available to the general public to reuse, read and modify. ↩︎
In particular, the Linux operating system and its Kernel-based Virtual Machine (KVM) module are two basic, essential, open-source building blocks upon which these hyperscalers are built. ↩︎
Just as most of today’s computing landscape, a witty reader might add ↩︎
For instance, Google Cloud does not rely on QEMU. In an attempt to reduce the attack surface of the platform, its designers decided to replace QEMU – which by default comes with a vast amount of generic features – with their own specialized user-space emulator, which is rumored to be called Vanadium. According to the Wikipedia article, Vanadium as as physical compound “is a hard, silvery-grey, malleable transition metal [...] rarely found in nature [...] [O]nce isolated artificially, the formation of an oxide layer [...] somewhat stabilizes the free metal against further oxidation.” Vanadium would surely be a fitting name for a user-space emulator... ↩︎
he fact that the hyperscalers’ foundations are closed source has several key consequences which go beyond the issues that Phyllome OS seeks to address. Individuals, private companies, and public entities increasingly rely on public clouds to run their most critical workloads. The core engine that propels those hyperscalers is made of computer code that no external party can audit, which means that the aforementioned critical workloads are ultimately not auditable. In other words, individuals, private companies and public entities are building their own essential digital infrastructures on top of black boxes. ↩︎
Crates are snippets of Rust code which provide certain functions or sets of functions. ↩︎
Hyperscalers tend to rely more on custom-made hardware, which might significantly raise the entry bar for new competitors in the future ↩︎