Comprehensive Overview of Server Virtualization Technologies

Server virtualization traces its roots to IBM mainframe z/VM and later PowerKVM/PowerVM, enabling hundreds of virtual machines on a single physical server.

Virtualization offers partitioning, isolation, and encapsulation, allowing multiple OS instances to share hardware while remaining independent.

CPU Virtualization

Three main approaches exist: full virtualization (VMM intercepts privileged instructions), paravirtualization (guest OS uses hypercalls), and hardware‑assisted virtualization (new CPU modes allow guest kernels to run directly on hardware).

Virtualization Software Architecture

Four primary models are described:

Hosted (type‑2) virtualization : hypervisor runs as an application on a host OS.

Bare‑metal (type‑1) virtualization : hypervisor runs directly on hardware, handling privileged instructions and scheduling.

OS‑level virtualization : the host OS itself partitions resources without a separate hypervisor (e.g., Docker, OpenVZ).

Hybrid virtualization : combines a host OS with a kernel‑level driver acting as a virtual hardware manager.

Memory Virtualization

Techniques include full memory virtualization with shadow page tables, paravirtual memory where the guest registers page tables with the VMM, and hardware‑assisted memory virtualization using Extended Page Tables (EPT) to translate guest physical addresses directly to host physical addresses.

I/O Virtualization

To avoid I/O bottlenecks, virtualization extends to network and storage devices. Full I/O virtualization emulates devices, paravirtual I/O uses front‑end/back‑end queues, and hardware‑assisted I/O employs Intel VT‑d, AMD IOMMU, and SR‑IOV, allowing direct device access or dedicated virtual functions.

Intel Hardware Support

VT‑x introduces root and non‑root operation modes for CPUs, enabling efficient execution of privileged instructions. VT‑d provides direct I/O access with DMA remapping and I/O page tables. VT‑c adds networking features such as VMDq (queue‑based packet classification) and VMDc (direct NIC access).

GPU Virtualization

GPU virtualization can be achieved via GPU passthrough (one VM per GPU), GPU sharing (GPU server with multiple clients), or full GPU virtualization where a physical GPU is split into multiple virtual GPUs (vGPUs) that can be concurrently used by several VMs.

Overall, bare‑metal and hybrid architectures combined with hardware‑assisted techniques represent the future direction for high‑performance server virtualization in cloud computing environments.