Cloud & Data Centre
CHERI can add fine-grained protection inside hosts, services, and infrastructure software where process or virtual-machine boundaries are too coarse.
Cloud platforms place valuable data beside large quantities of complex infrastructure code. Existing process, container, and virtual-machine boundaries remain important, but significant native code also runs inside each boundary.
Finer-grained least authority
CHERI can protect memory within an address space and support compartments that expose narrow interfaces. Potential targets include:
- network and storage data-processing paths;
- hypervisor, hardware-emulator, and device-model components;
- database extensions and native plug-ins;
- media, archive, and document parsers;
- language runtimes and native libraries;
- agents that hold credentials or broad host access.
A compartment can be given a bounded buffer and a small set of service capabilities rather than ambient access to the host process. If the component is compromised, its remaining authority is part of the design and can be tested.
Relationship to existing isolation
Capabilities do not replace tenant isolation, access control, virtual machines, containers, cryptography, or operational security. They can strengthen components inside those boundaries and create additional layers where a full process would be too expensive or awkward.
Memory-safe languages remain the preferred choice for new code where suitable. CHERI can protect legacy native code, unsafe portions of otherwise safe systems, and interfaces where multiple languages meet.
Evaluation questions
Measure the workloads that drive the service-level objective:
- What is the cost of capability-aware code and compartment transitions at realistic request rates?
- How do memory footprint, cache behaviour, latency tails, and throughput change?
- Which host services remain reachable after the target component is compromised?
- Can observability tools explain capability faults in production?
- How will images, orchestration, live updates, and fleet management support the new architecture?
- Does the selected platform have the operating-system and virtualisation maturity the deployment requires?
A focused pilot in an exposed parser, plug-in host, or infrastructure service can produce useful evidence without moving an entire cloud stack at once.