Split View: [가상화] 01. 가상화 기술의 기초: Type 1 vs Type 2 하이퍼바이저

[Virtualization] 01. Virtualization Fundamentals: Type 1 vs Type 2 Hypervisors

• Introduction
  • Why Virtualization Matters
• What Is a Hypervisor?
  • Type 1 Hypervisors (Bare-Metal)
  • Type 2 Hypervisors (Hosted)
  • KVM: A Hybrid Approach
• Virtualization Techniques Compared
  • Full Virtualization
  • Paravirtualization
  • Hardware-Assisted Virtualization
• Comparison of Three Virtualization Techniques
• Hands-On: Verifying KVM-Based Virtualization
• CPU Privilege Level (Ring) Architecture
• VM Exits and Performance
• Summary

Introduction

Virtualization is the technology of creating multiple isolated virtual environments on top of physical hardware. It enables running several operating systems simultaneously on a single server, maximizing hardware utilization and reducing infrastructure costs.

Why Virtualization Matters

• Hardware Consolidation: Boosts average CPU utilization from 15-20% to 60-80%
• Isolation: Each VM has its own kernel, filesystem, and network stack
• Flexible Provisioning: Deploy new server environments within minutes
• Snapshots and Migration: Easy disaster recovery and zero-downtime migration
• Cloud Foundation: AWS, Azure, and GCP all run on virtualization

What Is a Hypervisor?

A hypervisor is the software layer that creates and manages virtual machines (VMs). Also called a VMM (Virtual Machine Monitor), it abstracts physical resources (CPU, memory, storage, network) and distributes them to each VM.

Type 1 Hypervisors (Bare-Metal)

Type 1 hypervisors are installed directly on hardware. They operate without a host OS, resulting in lower overhead and better performance.

+-------------------------------------------+
|   VM 1      |   VM 2      |   VM 3        |
|  (Ubuntu)   |  (Windows)  |  (CentOS)     |
+-------------------------------------------+
|          Type 1 Hypervisor                 |
|     (ESXi / Hyper-V / Xen)                |
+-------------------------------------------+
|          Physical Hardware                 |
|   (CPU, RAM, Storage, NIC)                |
+-------------------------------------------+

Notable Type 1 Hypervisors:

Type 2 Hypervisors (Hosted)

Type 2 hypervisors run as regular applications on top of a host operating system.

+-------------------------------------------+
|   VM 1      |   VM 2      |   VM 3        |
|  (Ubuntu)   |  (Windows)  |  (Fedora)     |
+-------------------------------------------+
|          Type 2 Hypervisor                 |
|  (VirtualBox / VMware Workstation)        |
+-------------------------------------------+
|           Host OS (Windows/macOS/Linux)    |
+-------------------------------------------+
|          Physical Hardware                 |
|   (CPU, RAM, Storage, NIC)                |
+-------------------------------------------+

Notable Type 2 Hypervisors:

KVM: A Hybrid Approach

KVM (Kernel-based Virtual Machine) occupies a unique position. It operates as a Linux kernel module that transforms Linux itself into a Type 1 hypervisor.

+-------------------------------------------+
|   VM 1      |   VM 2      |   VM 3        |
|  (Ubuntu)   |  (Windows)  |  (Fedora)     |
+-------------------------------------------+
|     QEMU (Device Emulation / Management)  |
+-------------------------------------------+
|  Linux Kernel + KVM Module (Type 1-like)  |
+-------------------------------------------+
|          Physical Hardware                 |
|   (CPU + VT-x/AMD-V, RAM, Storage, NIC)  |
+-------------------------------------------+

• The kvm.ko module directly leverages CPU hardware virtualization extensions (VT-x/AMD-V)
• Guest CPU instructions execute directly on hardware (near-native performance)
• The Linux kernel handles process scheduling, memory management, and I/O
• QEMU provides device emulation and VM management
• Technically Type 1, but runs within a full Linux OS, hence classified as "hybrid"

Virtualization Techniques Compared

Full Virtualization

In full virtualization, the guest OS runs completely unmodified. The hypervisor intercepts and translates privileged instructions from the guest.

Binary Translation approach:

Guest OS (runs in Ring 1)
    |
    v  Attempts privileged instruction
    |
Hypervisor (Ring 0) traps it
    |
    v  Translates to safe instructions and executes
    |
Physical Hardware

• No guest OS modification needed (supports Windows, Linux, etc.)
• Performance overhead from binary translation
• VMware's original approach

Paravirtualization

In paravirtualization, the guest OS is aware it runs in a virtualized environment. Instead of privileged instructions, it uses hypercalls to directly request services from the hypervisor.

Guest OS (modified - includes hypercall interface)
    |
    v  Issues hypercall (direct communication instead of traps)
    |
Hypervisor (processes request)
    |
    v
Physical Hardware

• Requires guest OS kernel modification (Linux possible, standard Windows not)
• Lower overhead than full virtualization, better performance
• Xen's original approach; VirtIO drivers also leverage the paravirtualization concept
• VirtIO is the paravirtualization standard for I/O devices (network, storage)

Hardware-Assisted Virtualization

In 2005-2006, Intel and AMD introduced CPU-level virtualization support.

Key Technologies:

• Intel VT-x (2005): Introduced VMX root/non-root modes
• AMD-V (2006): SVM (Secure Virtual Machine) extensions
• Ring -1 Concept: Added a dedicated privilege level for the hypervisor
• VT-d / AMD-Vi: IOMMU-based DMA remapping, device passthrough support
• EPT / NPT: Extended Page Tables for hardware-accelerated memory virtualization

+--------------------------------------------------+
|  Guest OS (Ring 3: User, Ring 0: Kernel)          |
|  --> Runs in VMX non-root mode                     |
+--------------------------------------------------+
|  Hypervisor (Ring -1 / VMX root mode)              |
|  --> Only intervenes on VM Exit                    |
+--------------------------------------------------+
|  Hardware (VT-x/AMD-V, EPT/NPT, VT-d/AMD-Vi)     |
+--------------------------------------------------+

• No guest OS modification needed + near-native performance
• Performance varies based on VM Exit frequency
• Foundation of all modern x86 virtualization

Comparison of Three Virtualization Techniques

Hands-On: Verifying KVM-Based Virtualization

# Check if CPU supports virtualization
# Look for vmx (Intel VT-x) or svm (AMD-V) flags
grep -E '(vmx|svm)' /proc/cpuinfo | head -1

# Verify KVM module is loaded
lsmod | grep kvm

# Example output:
# kvm_intel            368640  0
# kvm                 1028096  1 kvm_intel

# Create a VM using libvirt
virt-install \
  --name test-vm \
  --ram 2048 \
  --vcpus 2 \
  --disk size=20 \
  --os-variant ubuntu22.04 \
  --cdrom /path/to/ubuntu-22.04.iso

CPU Privilege Level (Ring) Architecture

[Before HW Virtualization]           [After HW Virtualization]

Ring 3: User Applications            Ring 3: User Applications
Ring 2: (unused)                     Ring 2: (unused)
Ring 1: (Guest OS - full virt)       Ring 1: (unused)
Ring 0: Hypervisor                   Ring 0: Guest OS (non-root)
                                      Ring -1: Hypervisor (VMX root)

With hardware virtualization, the guest OS operates normally at Ring 0 while the hypervisor controls everything from a higher privilege level (Ring -1).

VM Exits and Performance

In hardware-assisted virtualization, when a guest performs certain operations, the CPU switches from VMX non-root to root mode. This is called a VM Exit.

Common VM Exit Triggers:

• I/O port access
• Certain MSR (Model-Specific Register) access
• Interrupt handling
• Page faults (without EPT)
• CPUID instruction execution

VM Exits cost hundreds to thousands of CPU cycles, so reducing their frequency is key to performance. Paravirtual drivers like VirtIO reduce VM Exits in the I/O path.

[VM Exit Processing Flow]

Guest Execution (non-root)
    |
    v  Executes sensitive instruction
    |
VM Exit (non-root -> root transition, ~hundreds of cycles)
    |
    v  Hypervisor processes the instruction
    |
VM Entry (root -> non-root transition)
    |
    v  Guest execution resumes

Summary

Virtualization is the foundation of modern cloud infrastructure. Type 1 hypervisors serve production environments while Type 2 serves desktop use cases. The advent of hardware virtualization support (VT-x/AMD-V) dramatically reduced performance overhead, enabling the cloud computing era we live in today.