The release of the Cloud Hypervisor 48.0 has been announced, developed based on components from the Rust-VMM project, which is supported by Intel, Alibaba, Amazon, Google, and Red Hat. Rust-VMM is written in Rust and allows for the creation of hypervisors tailored for specific tasks. Cloud Hypervisor is one such hypervisor that provides a high-level virtual machine monitor (VMM) running on top of KVM or MSHV and optimized for cloud system tasks. The project code is available under Apache 2.0 and BSD licenses. Initially created by Intel, the hypervisor was transitioned under the Linux Foundation's oversight, after which companies like Alibaba, AMD, Ampere, ARM, ByteDance, Cyberus Technology, Microsoft, and Tencent Cloud joined the development.
Cloud Hypervisor focuses on running modern Linux distributions using paravirtualized devices based on virtio. Key objectives include: high responsiveness, low memory consumption, high performance, simplified configuration, and reduced potential attack vectors. It features the ability to boot the OS kernel directly without intermediate bootloaders, allowing for very rapid virtual machine starts (with less than 100 ms passing before control is handed to user space applications).
Support for emulation is kept to a minimum, prioritizing paravirtualization. The project supports x86_64, AArch64, and RISC-V architectures. Guest systems supported include 64-bit builds of Linux and Windows 10/Windows Server 2019, and there is the capability to use Cloud Hypervisor for running containers in virtual machines via the Kata Containers toolkit. Live migration is supported. of virtual machines between servers and lifecycle management of virtual machines via HTTP API.
In the new version:
- Rules have been established for the use of AI tools in development, prohibiting the acceptance of code generated using large language models and AI assistants such as ChatGPT, Gemini, Claude, and GitHub Copilot. The reason cited is the desire to avoid ambiguity in compliance with licensing requirements (AI is trained on code under various licenses, and the generated code can potentially be interpreted as a derivative work). The ban on AI is also motivated by the intention to reduce the burden on maintainers, freeing them from having to parse meaningless corrections created by AI.
- Experimental support has been added for the fw_cfg (Firmware Configuration) device, which allows the host environment to pass data and settings to the guest system, such as virtual machine boot configuration, memory layout information, and ACPI tables.
- Experimental support has been added for the ivshmem device to enable multiple guest systems to share a common memory area.
- Support for booting RISC-V systems has been added through handover to firmware (Firmware Boot) instead of directly launching the OS kernel.
- For x86_64 hosts using the KVM hypervisor, the limit on the number of available virtual processors (vCPUs) has been increased from 254 to 8192.
- In the virtio-blk device, optimization has been implemented that utilizes batching and processing of asynchronous I/O requests. This change has improved the performance of operations with small block sizes (<=16KB).
- The execution of the pause operation has been accelerated. of virtual machines, which is particularly noticeable in virtual machines with a large number of vCPUs.
- Documentation has been added for running guest systems with Windows 11 under Cloud Hypervisor.
- Support for Intel SGX (Software Guard Extensions) technology has been discontinued.
Source: opennet.ru
