Toggle navigation

Having strong design opinions about manageability is one thing.

Bringing them to life in an approach that actually delivers on them is another. Especially when it comes to operating clouds at scale.
To keep cloud hardware fleets running efficiently—without the usual management pains—we focused on a few key concepts.

Pivoting from convenience to the pursuit of perfection

Antoine de Saint-Exupery famously said, “Perfection is achieved, not when there is nothing more to add, but when there is nothing left to take away.”

In HyperCloud, we’ve standardised our entire networking, storage, and compute hardware range on flexible building block appliances. We treat all hardware the same way and we’ve removed every piece of hardware and software that isn’t essential to smooth and secure operations.

We’ve even standardised our operating system across compute, storage, and management layers with a custom GNU/Linux OS layer that runs on all HyperCloud host machines. Again, our approach is minimalistic, eliminating every line of unnecessary code to reduce the attack surface and maintenance burdens and focusing on the essentials of a “base OS layer to rule them all.”

Employing strict controls on both the hardware platform and software images in HyperCloud enables operators to rely on standardised and deterministic bare metal images. You can simply build once and ship everywhere, including compute, storage, and interconnect nodes. Tasks like upgrading nodes become as simple as rebooting.

Bare-metal provisioning

Today’s popular private cloud approaches essentially pursue statelessness as a religion at the application and container orchestration layers while ignoring it closer to the metal. That’s a problem because with state, the complexity and inconsistency that makes systems difficult to manage at scale is lethal. In fact, service failures, delayed recovery, and discrete points of failure are par for the course in stateful systems; carrying any cluster outside of the control plane is usually a mistake when managing infrastructure at scale. Especially on the metal.

By standardising on a single platform and maintaining our own out-of-band hardware and software, users can provision HyperCloud from a distributed control plane that carries all the state and nodes. To perform upgrades and lifecycle management, you simply update the control plane and reboot stateless nodes. You can also easily netboot the rest of the fleet from the control plane and nodes don’t even need a root disk to boot.

HyperCloud delivers stateless bring-up through a set of stateful control planes that both act as the interconnect nodes in each rack and look after the bring-up of stateless nodes. Ultimately, HyperCloud enables you to provision, commission, and decommission hardware on the fly without having to worry about the state of servers.

Calculated placement

Typical cloud approaches categorise nodes in management, controller, and similar roles rather than placing data and I/O where it needs to be based on fluctuating workloads. Unsurprisingly, the outcomes include bottlenecks and points of failure.

HyperCloud is built on a cluster-aware architecture that uses hashing algorithms to help place data and I/O where it makes the most sense from an efficiency standpoint with minimal effort.

Centralised management

Clouds based on software defined infrastructure and even integrated infrastructures rely on a confounding mixture of management systems. HyperCloud centralises management—from out -of-band-controls to storage subsystems to container orchestration—in a unified control plane. It’s all based on a standardised API that provides complete oversight and management across firmware and appliances.