Live Partition Mobility (LPM) in IBM Power Systems (part of PowerVM) lets you move a running logical partition (LPAR) from one physical server to another without downtime. Internally, it’s a carefully orchestrated mix of hypervisor coordination, memory copying, and state transfer.

🔧 Internal Working of Live Partition Mobility

1. Pre-Migration Validation

Before anything moves:

• Source and target systems are checked for compatibility:
  • Same processor family (e.g., POWER8 → POWER10 allowed with compatibility mode)
  • Shared storage access (SAN, NPIV, or virtual SCSI)
  • Network virtualization readiness via VIOS
• Resources (CPU, memory) must be available on target

👉 This step prevents failure mid-migration.

2. Hypervisor Coordination

The Power Hypervisor (PHYP)—the firmware layer in IBM Power Systems—controls the process:

• Creates a target LPAR shell on destination server
• Establishes a secure communication channel between source and target
• Coordinates with Virtual I/O Servers (VIOS) on both sides

3. Memory Pre-Copy Phase (Key Mechanism)

This is the core of LPM:

• Memory pages are copied while the LPAR is still running
• Called iterative pre-copy:
  1. Copy all memory pages
  2. Track pages modified during copy (“dirty pages”)
  3. Re-copy only dirty pages
  4. Repeat until remaining dirty set is very small

👉 This ensures minimal downtime later.

4. CPU and State Synchronization

Once memory delta is small:

• LPAR is briefly paused (milliseconds)
• CPU register state, processor context, and remaining memory pages are transferred
• Device states (virtual NICs, disks) are synchronized

5. Network Virtualization Handling

Networking continues seamlessly using:

• Virtual Ethernet adapters via VIOS
• MAC/IP identity preserved

The network session does not drop because:

• Same virtual network configuration exists on target
• External switches see no change

6. Storage Continuity

Disk access is uninterrupted because:

• Storage is shared (SAN / NPIV / vSCSI)
• No disk data is copied during migration
• Only I/O mappings shift to target VIOS

7. Final Switchover (Cutover Phase)

• Source LPAR stops execution
• Target LPAR resumes instantly with:
  • Same memory state
  • Same CPU state
  • Same network identity

👉 Downtime is typically < 1 second

8. Cleanup Phase

• Source LPAR is removed
• Resources are freed
• VIOS mappings on source are cleaned up

⚙️ Key Internal Concepts

🔹 1. Active Memory Sharing Awareness

If AMS is used:

• Hypervisor ensures correct page ownership transfer
• Shared memory pools are updated dynamically

🔹 2. Dirty Page Tracking

• Hardware-assisted memory tracking
• Uses page tables to detect writes during migration

🔹 3. Compression & Optimization

• Memory pages may be compressed before transfer
• Reduces network bandwidth usage

🔹 4. Secure Transfer Channel

• Encrypted communication between systems
• Prevents interception of in-memory data

📊 Migration Timeline (Simplified)

Start → Pre-check → Memory copy → Dirty page iterations →
Pause → Final sync → Resume on target → Cleanup

🚀 Why LPM is Powerful

• Zero downtime maintenance
• Load balancing across servers
• Hardware upgrades without service interruption
• Disaster avoidance (move workloads proactively)

⚠️ Limitations

• Requires shared storage
• High network bandwidth recommended
• Not suitable for extremely memory-write-intensive workloads (may delay convergence)