Deep Dive: Hyper-V 架构与工作机制全解析

Topic: Windows Hyper-V Architecture & Working Mechanisms
Category: Virtualization / Infrastructure
Level: 中级 → 高级
Last Updated: 2026-03-17

1. 概述 (Overview)

Hyper-V 是 Microsoft 的企业级虚拟化平台，自 Windows Server 2008 引入以来持续演进。它采用 Type-1 (Bare Metal) Hypervisor 架构——Hypervisor 直接运行在硬件之上，而非运行在操作系统之中（这与 VMware Workstation / VirtualBox 等 Type-2 架构有本质区别）。

Hyper-V 的核心使命是：将一台物理服务器的 CPU、内存、存储、网络资源虚拟化，分配给多个隔离的虚拟机 (VM)，同时保持接近原生的性能。它是 Windows Server Failover Clustering、Azure Local (原 Azure Stack HCI)、Azure 公有云的底层虚拟化引擎。

理解 Hyper-V 的内部机制，对于排查 VM 性能问题、Live Migration 失败、存储 IO 瓶颈、网络延迟等场景至关重要。

2. 核心概念 (Core Concepts)

2.1 Type-1 Hypervisor

┌─────────────────────────────────────────────────┐
│              虚拟机 (Child Partitions)            │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐      │
│  │   VM 1   │  │   VM 2   │  │   VM 3   │      │
│  │  (VSC)   │  │  (VSC)   │  │  (VSC)   │      │
│  └────┬─────┘  └────┬─────┘  └────┬─────┘      │
│       │              │              │    VMBus    │
│  ┌────┴──────────────┴──────────────┴─────┐      │
│  │         Root Partition (Host OS)        │      │
│  │   VMMS / VMWP / VSP / Device Drivers   │      │
│  └────────────────┬───────────────────────┘      │
├───────────────────┼──────────────────────────────┤
│            Hypervisor (hvix64.exe)                │
├───────────────────┼──────────────────────────────┤
│         Physical Hardware (CPU/RAM/NIC/Disk)      │
└─────────────────────────────────────────────────┘

关键理解：安装 Hyper-V Role 后，原来的 Windows OS 不再直接运行在硬件上，而是被”降级”为 Root Partition（管理分区），Hypervisor 才是真正的”老大”。

类比：把 Hypervisor 想象成酒店大堂经理，Root Partition 是前台接待（可以调度资源），Child Partition 是各个房间的住客（每个 VM）。前台虽然权限最大，但也得听大堂经理的。

2.2 Partition（分区）

• Root Partition 是 Hypervisor 创建的第一个分区
• 所有 Child Partition 由 Root Partition 的 VMMS 服务创建和管理
• Root Partition 对物理设备有完全控制权（中断处理、DMA 访问等）

2.3 VMBus（虚拟机总线）

VMBus 是 Root Partition 和 Child Partition 之间的高速通信通道，基于共享内存实现。它替代了传统设备模拟的低效方式。

Child Partition              Root Partition
┌─────────────┐              ┌─────────────┐
│    Guest OS  │              │    Host OS   │
│   ┌───────┐  │    VMBus     │  ┌────────┐  │
│   │  VSC  │──┼──────────────┼──│  VSP   │  │
│   │(Client)│  │  共享内存     │  │(Provider)│ │
│   └───────┘  │              │  └────┬───┘  │
└─────────────┘              │       │       │
                              │  Physical HW │
                              └─────────────┘

2.4 VSP / VSC（虚拟化服务提供者/消费者）

常见 VSP/VSC 对：

• 网络 VSP/VSC → 虚拟网卡 (Synthetic NIC)
• 存储 VSP/VSC → 虚拟 SCSI 控制器
• 视频 VSP/VSC → 虚拟显卡
• HID VSP/VSC → 虚拟键盘/鼠标

2.5 Enlightenments（光照优化/启蒙）

Enlightenments 是 Guest OS 内核中的虚拟化感知优化。当 Guest OS 检测到自己运行在 Hyper-V 上时，会启用一系列优化路径来提升性能：

• Hypercall：Guest OS 直接通过 Hypercall 接口与 Hypervisor 通信（类似系统调用）
• 合成中断：取代传统 PIC/APIC 模拟，减少虚拟化开销
• TSC 同步：时间戳计数器同步，避免 VM 时间漂移
• 自旋锁优化：多 vCPU 场景下避免无效自旋等待

关键点：这就是为什么安装 Integration Services（集成服务） 如此重要——它包含 VSC 驱动和 Enlightenment 组件。没有它们，VM 的所有 IO 都要走低效的设备模拟路径。

2.6 VMMS 与 VMWP

vmms.exe (1个)
  ├── vmwp.exe (VM-1 的 Worker Process)
  ├── vmwp.exe (VM-2 的 Worker Process)
  └── vmwp.exe (VM-3 的 Worker Process)

排查意义：如果某个 VM 卡死，可以在 Task Manager 中找到对应的 vmwp.exe 进程。通过 VM 的 GUID 可以关联到具体哪个 vmwp.exe。

3. 工作原理 (How It Works)

3.1 VM 代际：Gen1 vs Gen2

Gen2 VM 的关键优势：所有设备都是 Synthetic（合成）设备通过 VMBus 通信，无需任何设备模拟（无 IDE、无 PS/2、无 PIC），因此性能更好、攻击面更小。

3.2 存储 IO 路径

存储 IO 需要经过四层：

┌───────────────────────────────────────┐
│ 1. Guest Storage Stack                │
│    应用 → 文件系统 → 存储驱动(VSC)      │
├───────────────────────────────────────┤
│ 2. Host Virtualization Layer          │
│    VMBus → VSP → 设备模拟              │
├───────────────────────────────────────┤
│ 3. Host Storage Stack                 │
│    NTFS/ReFS → StorPort → 小端口驱动    │
├───────────────────────────────────────┤
│ 4. Physical Disk                      │
│    HDD / SSD / NVMe                   │
└───────────────────────────────────────┘

虚拟硬盘格式对比：

磁盘类型对比：

3.3 Checkpoint 机制（检查点/快照）

Checkpoint 创建时，Hyper-V 不会复制整个磁盘，而是：

1. 冻结当前 VHDX，使其变为只读
2. 创建一个 AVHDX（差异磁盘） 文件
3. 所有后续写操作都写入 AVHDX
4. 可选地保存内存状态（Standard Checkpoint）

Base.vhdx (只读)
  └── Checkpoint1.avhdx (只读)
       └── Checkpoint2.avhdx (当前活跃)

两种 Checkpoint 类型：

AVHDX Merge（合并）过程：

当删除 Checkpoint 时，Hyper-V 将 AVHDX 中的变更合并回父盘：

• VM 运行中：后台异步 Merge（Live Merge），可能需要较长时间
• VM 关闭后：在下次启动时或关机过程中完成 Merge

⚠️ 常见故障：AVHDX 链过长（>3层）会严重影响 IO 性能；如果 Merge 失败（如空间不足），可能导致 VM 无法启动。

3.4 Dynamic Memory（动态内存）

Dynamic Memory 允许 Hyper-V 在多个 VM 之间智能分配和回收内存：

┌──────────────────────────────────────────┐
│            Memory Balancer               │
│  (监控所有 VM 内存压力，动态分配/回收)      │
├──────────┬──────────┬────────────────────┤
│  VM-1    │  VM-2    │  VM-3              │
│ Min: 1GB │ Min: 2GB │ Min: 512MB         │
│ Start:2GB│ Start:4GB│ Start: 1GB         │
│ Max: 8GB │ Max:16GB │ Max: 4GB           │
│ 当前: 3GB│ 当前:12GB│ 当前: 1GB          │
└──────────┴──────────┴────────────────────┘

三个关键参数：

Smart Paging（智能分页）：

解决一个极端场景：VM 当前只有 Minimum Memory，但需要重启（重启需要 Startup Memory 大小）。此时：

1. 如果 Host 物理内存不够提供 Startup Memory
2. Hyper-V 在磁盘上创建临时分页文件给 VM 用
3. VM 启动完成后自动释放

Smart Paging 是”最后手段”，磁盘远慢于内存，只在内存极度紧张时使用。

Balloon Driver 机制：

Hyper-V Integration Services 中包含一个 Balloon Driver（气球驱动）：

• 当 Host 需要回收 VM 内存时，Balloon Driver 在 Guest 内部”膨胀”，占用 Guest 内存页
• Guest OS 看到可用内存减少，会减少自己的内存使用
• Hypervisor 然后安全地将这些内存页分配给其他 VM
• 反向操作（”放气”）则释放内存给 Guest

3.5 虚拟网络 (Virtual Networking)

虚拟交换机类型

网络性能优化技术

┌──────────────────────────────────────────┐
│        高性能网络技术栈                    │
├──────────────────────────────────────────┤
│  SR-IOV      │ 绕过 VMBus，VM 直接      │
│              │ 访问物理 NIC 的虚拟功能    │
│              │ → 最低延迟、最高吞吐       │
├──────────────┼───────────────────────────┤
│  VMQ         │ 硬件队列分离，每个 VM      │
│              │ 有独立的 NIC 接收队列      │
│              │ → 减少 CPU 中断合并开销    │
├──────────────┼───────────────────────────┤
│  vRSS        │ VM 内部多核接收分发        │
│              │ → 突破单核网络处理瓶颈      │
├──────────────┼───────────────────────────┤
│  SET         │ Switch Embedded Teaming   │
│              │ 集成到 vSwitch 中的        │
│              │ NIC 绑定（替代 LBFO）      │
└──────────────┴───────────────────────────┘

关键变化 (Server 2022+)：LBFO NIC Teaming 已被弃用用于 Hyper-V vSwitch，必须使用 SET (Switch Embedded Teaming)。

SR-IOV 详解

普通路径:
  VM → VSC → VMBus → VSP → Host NIC Driver → Physical NIC

SR-IOV 路径:
  VM → VF Driver → Physical NIC (直通!)
  (VMBus 仅用于控制平面，数据平面完全绕过)

SR-IOV 需要：硬件支持 + BIOS/UEFI 开启 + NIC 驱动支持 + VM 使用 Synthetic NIC 并开启 SR-IOV。Live Migration 时 SR-IOV 会自动降级为 Synthetic 路径，迁移完成后恢复。

3.6 Live Migration（实时迁移）

Live Migration 是在不中断 VM 服务的情况下将运行中的 VM 从一台 Host 迁移到另一台。

完整过程

sequenceDiagram
    participant Src as Source Host
    participant Dst as Destination Host
    participant VM as Running VM
    
    Note over Src,Dst: Phase 1 - Setup
    Src->>Dst: Compatibility check (CPU/Network/Storage)
    Dst->>Src: Ready, resources reserved
    
    Note over Src,Dst: Phase 2 - Pre-copy (Memory Transfer)
    Src->>Dst: Copy ALL memory pages (Round 1)
    Note over VM: VM continues running, dirtying pages
    Src->>Dst: Copy DIRTY pages only (Round 2)
    Src->>Dst: Copy DIRTY pages only (Round 3)
    Note over Src,Dst: Rounds continue until dirty set is small enough
    
    Note over Src,Dst: Phase 3 - Blackout (Switchover)
    Src->>VM: PAUSE VM
    Src->>Dst: Transfer final dirty pages + CPU state + device state
    Dst->>VM: RESUME VM on Destination
    
    Note over Src,Dst: Phase 4 - Cleanup
    Dst-->>Src: Migration complete, release source resources

关键术语

性能选项

认证方式

Quick Migration vs Live Migration

3.7 Hyper-V Replica（副本）

Hyper-V Replica 提供异步复制用于灾难恢复（DR）：

┌──────────────┐          ┌──────────────┐
│ Primary Site  │  复制     │ Replica Site  │
│              │ ──────→  │              │
│   VM (运行)   │          │  VM (关闭)    │
│              │  HTTP/   │              │
│              │  HTTPS   │              │
└──────────────┘          └──────────────┘

工作流程：

1. Initial Replication：完整复制 VM 的 VHD/VHDX 到目标站点（可通过网络或物理介质）
2. Delta Replication：定期（30秒/5分钟/15分钟）传输变化块
3. Change Tracking：使用 Resilient Change Tracking (RCT) 在 VHDX 级别追踪变化
4. Recovery Points：可保留多个历史恢复点

三种 Failover 类型：

4. 关键配置与参数 (Key Configurations)

4.1 Hyper-V Host 配置

4.2 VM 配置

4.3 性能监控关键计数器

5. 常见问题与排查 (Common Issues & Troubleshooting)

问题 1: VM 无法启动

症状： Hyper-V Manager 中 Start VM 失败，状态卡在 Starting 或直接报错

常见原因及排查：

关键事件日志路径：

• Applications and Services Logs > Microsoft > Windows > Hyper-V-VMMS > Admin
• Applications and Services Logs > Microsoft > Windows > Hyper-V-Worker > Admin

问题 2: Live Migration 失败

症状： Migration 过程中报错回退，或 VM 在目标 Host 不可达

排查思路：

1. 兼容性检查：Compare-VM 或检查 Event Log 中的 Compatibility 错误
2. CPU 兼容性：不同代 CPU 需开启 “Processor Compatibility Mode”
3. 网络连通性：确认 LM 专用网络双向连通
4. 认证问题：Kerberos 需要 Constrained Delegation 配置；CredSSP 需要先登录源主机
5. 存储可达性：确认两台 Host 都能访问 VM 的存储路径

关键命令：

# 测试 Live Migration 兼容性
Compare-VM -Name "TestVM" -DestinationHost "HV-Node02"

# 查看 Live Migration 配置
Get-VMHost | Select-Object VirtualMachineMigration*

# 测试网络连通
Test-NetConnection -ComputerName HV-Node02 -Port 6600

问题 3: VM 性能缓慢

症状： VM 内应用响应慢，Guest 内 CPU/内存/磁盘指标异常

排查层次：

1. 检查 Host 层面是否过载
   → CPU: Physical Processor % Total Run Time
   → Memory: Available Memory < 2GB?
   → Disk: Host 磁盘队列长度

2. 检查 VM 层面虚拟化开销
   → vCPU Wait Time (CPU 超分?)
   → Dynamic Memory 是否频繁 Balloon/Smart Page
   → AVHDX 链是否过长

3. 检查 Guest 内部
   → Guest OS 自身问题(驱动、服务、应用)
   → Integration Services 版本是否最新

问题 4: VMMS/VMWP 服务崩溃

症状： vmms.exe 或 vmwp.exe 意外终止，VM 变为 Critical 状态

排查方法：

1. 检查 Windows Error Reporting (WER)：C:\ProgramData\Microsoft\Windows\WER\ReportQueue
2. 收集 Crash Dump：配置 WER 生成 Full Dump
3. 检查是否有第三方安全软件/备份软件干扰
4. 检查最近的 Windows Update 是否引入了已知问题

6. 实战经验 (Practical Tips)

最佳实践

• 始终使用 Gen2 VM 除非有明确的兼容性需求
• 生产 VM 用 Fixed VHDX：避免 Dynamic Disk 的扩展开销
• Checkpoint ≠ Backup：不要用 Checkpoint 替代正规备份。AVHDX 链过长会严重影响性能
• 专用网络分离：Management、Live Migration、Storage、VM Traffic 各用独立网络/VLAN
• NUMA 感知：高性能 VM (如 SQL) 将 vCPU 和内存配置在单个 NUMA 节点内
• Integration Services 保持最新：影响 VSC 驱动性能和功能

常见误区

安全注意

• Shielded VM 用于高安全需求场景（需要 HGS 基础设施）
• Secure Boot 对 Gen2 VM 默认启用，确认 Linux VM 使用正确的证书模板
• BitLocker + vTPM 可加密 VM 磁盘
• PowerShell Direct 可在不依赖网络的情况下管理 VM（通过 VMBus）

7. 与相关技术的对比 (Comparison)

8. 参考资料 (References)

• Hyper-V Architecture — 官方架构文档，详细描述 VSP/VSC/VMBus 机制
• Hyper-V Features and Terminology — 全面的功能和术语参考
• Hyper-V Overview — 官方概述，涵盖所有企业功能
• Should I Create a Gen 1 or Gen 2 VM? — Gen1 vs Gen2 完整对比
• Live Migration Overview — Live Migration 官方文档
• Hyper-V Replica Setup — Replica 配置和工作原理
• Dynamic Memory for Hyper-V — 动态内存官方指南

English Version

Deep Dive: Hyper-V Architecture & Working Mechanisms

Topic: Windows Hyper-V Architecture & Working Mechanisms
Category: Virtualization / Infrastructure
Level: Intermediate → Advanced
Last Updated: 2026-03-17

1. Overview

Hyper-V is Microsoft’s enterprise-grade virtualization platform, first introduced in Windows Server 2008 and continuously evolved since. It employs a Type-1 (Bare Metal) Hypervisor architecture — the hypervisor runs directly on hardware, not inside an operating system (fundamentally different from Type-2 hypervisors like VMware Workstation or VirtualBox).

Hyper-V’s core mission: virtualize a physical server’s CPU, memory, storage, and network resources, allocating them to multiple isolated virtual machines (VMs) while maintaining near-native performance. It serves as the underlying virtualization engine for Windows Server Failover Clustering, Azure Local (formerly Azure Stack HCI), and the Azure public cloud.

Understanding Hyper-V internals is crucial for troubleshooting VM performance issues, live migration failures, storage IO bottlenecks, and network latency scenarios.

2. Core Concepts

2.1 Type-1 Hypervisor

┌─────────────────────────────────────────────────┐
│              Virtual Machines (Child Partitions)  │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐      │
│  │   VM 1   │  │   VM 2   │  │   VM 3   │      │
│  │  (VSC)   │  │  (VSC)   │  │  (VSC)   │      │
│  └────┬─────┘  └────┬─────┘  └────┬─────┘      │
│       │              │              │    VMBus    │
│  ┌────┴──────────────┴──────────────┴─────┐      │
│  │         Root Partition (Host OS)        │      │
│  │   VMMS / VMWP / VSP / Device Drivers   │      │
│  └────────────────┬───────────────────────┘      │
├───────────────────┼──────────────────────────────┤
│            Hypervisor (hvix64.exe)                │
├───────────────────┼──────────────────────────────┤
│         Physical Hardware (CPU/RAM/NIC/Disk)      │
└─────────────────────────────────────────────────┘

Key insight: After installing the Hyper-V role, the original Windows OS no longer runs directly on hardware. It becomes the Root Partition (management partition), while the Hypervisor becomes the true “boss.”

Analogy: Think of the Hypervisor as a hotel general manager, the Root Partition as the front desk (can dispatch resources), and Child Partitions as guest rooms (each VM). The front desk has the most authority, but still answers to the general manager.

2.2 Partitions

• Root Partition is the first partition created by the Hypervisor
• All Child Partitions are created and managed by VMMS in the Root Partition
• Root Partition has full control over physical devices (interrupt handling, DMA access, etc.)

2.3 VMBus (Virtual Machine Bus)

VMBus is the high-speed communication channel between Root and Child Partitions, implemented via shared memory. It replaces the inefficient traditional device emulation approach.

Child Partition              Root Partition
┌─────────────┐              ┌─────────────┐
│    Guest OS  │              │    Host OS   │
│   ┌───────┐  │    VMBus     │  ┌────────┐  │
│   │  VSC  │──┼──────────────┼──│  VSP   │  │
│   │(Client)│  │ Shared Memory│  │(Provider)│ │
│   └───────┘  │              │  └────┬───┘  │
└─────────────┘              │       │       │
                              │  Physical HW │
                              └─────────────┘

2.4 VSP / VSC

Common VSP/VSC pairs:

• Network VSP/VSC → Synthetic NIC
• Storage VSP/VSC → Virtual SCSI Controller
• Video VSP/VSC → Virtual GPU
• HID VSP/VSC → Virtual Keyboard/Mouse

2.5 Enlightenments

Enlightenments are virtualization-aware optimizations in the Guest OS kernel. When the Guest OS detects it’s running on Hyper-V, it enables optimized code paths:

• Hypercalls: Guest OS communicates directly with the Hypervisor (similar to system calls)
• Synthetic Interrupts: Replace traditional PIC/APIC emulation, reducing virtualization overhead
• TSC Synchronization: Time Stamp Counter sync to prevent VM time drift
• Spinlock Optimization: Avoid wasteful spinning in multi-vCPU scenarios

Key point: This is why installing Integration Services is critical — they contain VSC drivers and Enlightenment components. Without them, all VM IO goes through the slow device emulation path.

2.6 VMMS and VMWP

vmms.exe (1 instance)
  ├── vmwp.exe (Worker for VM-1)
  ├── vmwp.exe (Worker for VM-2)
  └── vmwp.exe (Worker for VM-3)

Troubleshooting significance: If a specific VM hangs, find its corresponding vmwp.exe in Task Manager. The VM’s GUID maps to a specific vmwp.exe instance.

3. How It Works

3.1 VM Generations: Gen1 vs Gen2

Gen2 key advantage: All devices are Synthetic (through VMBus), no device emulation needed (no IDE, no PS/2, no PIC), resulting in better performance and smaller attack surface.

3.2 Storage IO Path

Storage IO traverses four layers:

┌───────────────────────────────────────┐
│ 1. Guest Storage Stack                │
│    App → File System → Storage VSC    │
├───────────────────────────────────────┤
│ 2. Host Virtualization Layer          │
│    VMBus → VSP → Device Emulation     │
├───────────────────────────────────────┤
│ 3. Host Storage Stack                 │
│    NTFS/ReFS → StorPort → Miniport    │
├───────────────────────────────────────┤
│ 4. Physical Disk                      │
│    HDD / SSD / NVMe                   │
└───────────────────────────────────────┘

Virtual Hard Disk Format Comparison:

Disk Type Comparison:

3.3 Checkpoint Mechanism

When a checkpoint is created, Hyper-V does NOT copy the entire disk. Instead:

1. Freezes the current VHDX, making it read-only
2. Creates an AVHDX (differencing disk) file
3. All subsequent writes go to the AVHDX
4. Optionally saves memory state (Standard Checkpoint)

Base.vhdx (read-only)
  └── Checkpoint1.avhdx (read-only)
       └── Checkpoint2.avhdx (currently active)

Two Checkpoint Types:

AVHDX Merge Process:

When deleting a checkpoint, Hyper-V merges AVHDX changes back into the parent:

• VM running: Background async merge (Live Merge), may take a long time
• VM powered off: Merge completes during shutdown or next startup

⚠️ Common issue: Long AVHDX chains (>3 levels) severely impact IO performance; failed merges (e.g., insufficient space) can prevent VM from starting.

3.4 Dynamic Memory

Dynamic Memory enables Hyper-V to intelligently allocate and reclaim memory across VMs:

Three Key Parameters:

Smart Paging:

Addresses an edge case: VM currently has only Minimum Memory but needs to restart (restart requires Startup Memory). When the host can’t provide enough physical RAM:

1. Hyper-V creates a temporary paging file on disk for the VM
2. VM boots using disk-backed memory (slow but functional)
3. Automatically released once memory pressure eases

Smart Paging is a “last resort” — disk is far slower than RAM.

Balloon Driver Mechanism:

Integration Services includes a Balloon Driver:

• When the host needs to reclaim VM memory, the Balloon Driver “inflates” inside the guest, consuming guest memory pages
• Guest OS sees reduced available memory and decreases its usage
• Hypervisor safely reallocates those pages to other VMs
• Reverse operation (“deflate”) releases memory back to the guest

3.5 Virtual Networking

Virtual Switch Types

Network Performance Technologies

Key change (Server 2022+): LBFO NIC Teaming is deprecated for Hyper-V vSwitch. Must use SET (Switch Embedded Teaming).

SR-IOV Details

Normal path:
  VM → VSC → VMBus → VSP → Host NIC Driver → Physical NIC

SR-IOV path:
  VM → VF Driver → Physical NIC (direct!)
  (VMBus only for control plane, data plane completely bypassed)

During Live Migration, SR-IOV automatically falls back to the Synthetic path, then re-enables after migration completes.

3.6 Live Migration

Live Migration moves a running VM from one host to another without service interruption.

Complete Process

sequenceDiagram
    participant Src as Source Host
    participant Dst as Destination Host
    participant VM as Running VM
    
    Note over Src,Dst: Phase 1 - Setup
    Src->>Dst: Compatibility check (CPU/Network/Storage)
    Dst->>Src: Ready, resources reserved
    
    Note over Src,Dst: Phase 2 - Pre-copy (Memory Transfer)
    Src->>Dst: Copy ALL memory pages (Round 1)
    Note over VM: VM continues running, dirtying pages
    Src->>Dst: Copy DIRTY pages only (Round 2)
    Src->>Dst: Copy DIRTY pages only (Round 3)
    Note over Src,Dst: Rounds continue until dirty set is small enough
    
    Note over Src,Dst: Phase 3 - Blackout (Switchover)
    Src->>VM: PAUSE VM
    Src->>Dst: Transfer final dirty pages + CPU state + device state
    Dst->>VM: RESUME VM on Destination
    
    Note over Src,Dst: Phase 4 - Cleanup
    Dst-->>Src: Migration complete, release source resources

Key Terminology

Performance Options

Quick Migration vs Live Migration

3.7 Hyper-V Replica

Hyper-V Replica provides asynchronous replication for disaster recovery:

┌──────────────┐          ┌──────────────┐
│ Primary Site  │  Repl.   │ Replica Site  │
│              │ ──────→  │              │
│   VM (Online) │          │  VM (Off)     │
│              │  HTTP/   │              │
│              │  HTTPS   │              │
└──────────────┘          └──────────────┘

Workflow:

1. Initial Replication: Full VM VHD/VHDX copy to target site (via network or physical media)
2. Delta Replication: Periodic (30s/5min/15min) change block transfers
3. Change Tracking: Uses Resilient Change Tracking (RCT) at VHDX level
4. Recovery Points: Can maintain multiple historical recovery points

Three Failover Types:

4. Key Configurations

4.1 Hyper-V Host

4.2 Performance Monitoring Counters

5. Common Issues & Troubleshooting

Issue 1: VM Won’t Start

Common causes:

Issue 2: Live Migration Failures

Troubleshooting steps:

1. Compatibility: Compare-VM or check Event Log compatibility errors
2. CPU compatibility: Different CPU generations require “Processor Compatibility Mode”
3. Network: Verify LM dedicated network bidirectional connectivity
4. Authentication: Kerberos needs Constrained Delegation; CredSSP needs source host login
5. Storage accessibility: Both hosts must access VM storage paths

Issue 3: VM Performance Degradation

Layered troubleshooting:

1. Host level: CPU overload? Memory pressure? Disk queue length?
2. Virtualization overhead: vCPU Wait Time? Frequent balloon/smart paging? Long AVHDX chains?
3. Guest internal: Guest OS issues? Integration Services up-to-date?

Issue 4: VMMS/VMWP Crashes

1. Check Windows Error Reporting: C:\ProgramData\Microsoft\Windows\WER\ReportQueue
2. Collect crash dumps
3. Check for third-party security/backup software interference
4. Check recent Windows Updates for known issues

6. Practical Tips

Best Practices

• Always use Gen2 VMs unless explicit compatibility needs
• Fixed VHDX for production — avoid Dynamic Disk expansion overhead
• Checkpoints ≠ Backups — don’t substitute checkpoints for proper backups
• Dedicated network separation — Management, Live Migration, Storage, VM Traffic on separate networks/VLANs
• NUMA-aware configuration — for high-performance VMs (e.g., SQL), keep vCPU and memory within single NUMA node
• Keep Integration Services current — impacts VSC/Enlightenment performance

Common Misconceptions

7. Comparison with Related Technologies

8. References

• Hyper-V Architecture — Official architecture document detailing VSP/VSC/VMBus mechanisms
• Hyper-V Features and Terminology — Comprehensive feature and terminology reference
• Hyper-V Overview — Official overview covering all enterprise capabilities
• Should I Create a Gen 1 or Gen 2 VM? — Complete Gen1 vs Gen2 comparison
• Live Migration Overview — Official Live Migration documentation
• Hyper-V Replica Setup — Replica configuration and working principles
• Dynamic Memory for Hyper-V — Official Dynamic Memory guide