Live Partition Mobility (LPM) in IBM PowerVM keeps memory consistent by using a controlled, iterative copy + synchronization mechanism while the LPAR is still running.

The goal is simple:
👉 Move a running system without losing or corrupting any memory state

🚀 Core Idea

Memory is not copied just once. Instead, PowerVM uses:

1. Pre-copy (iterative memory transfer)
2. Dirty page tracking
3. Final synchronization (brief pause)

👉 Ensures source and target memory stay logically identical

⚙️ Step-by-Step Memory Consistency Process

1. Initial Memory Copy (Pre-Copy Phase)

• Hypervisor starts copying memory pages:

  Source → Target system
  

• LPAR continues running normally

👉 Challenge:

• Some pages change during copying

2. Dirty Page Tracking

• Hypervisor tracks which pages are modified (“dirty” pages)

👉 Mechanism:

• Uses hardware-assisted page tracking
• Marks pages written during migration

3. Iterative Re-Copy

• Only dirty pages are copied again
• This cycle repeats multiple times

👉 Result:

• Number of dirty pages gradually reduces

4. Stop-and-Sync Phase (Critical Step)

• LPAR is briefly paused (milliseconds)
• Final dirty pages are copied
• CPU register state + memory state synchronized

👉 Now:

• Target has a fully consistent snapshot

5. Resume on Target

• LPAR resumes execution on target system
• Source instance is stopped

👉 No data loss, no corruption

🔑 Key Techniques Ensuring Consistency

🔹 Write Tracking (Dirty Bit Mechanism)

• Detects memory changes in real time
• Ensures no updates are missed

🔹 Copy-on-Write Semantics

• If a page changes during transfer:
  • It is re-copied
• Prevents stale data at destination

🔹 CPU State Synchronization

• Registers, program counters, cache state are transferred

👉 Ensures execution resumes exactly where it left off

🔹 Network Buffer & I/O State Handling

• In-flight I/O is coordinated via:
  • Virtual I/O Server (VIOS)
• Ensures:
  • No packet/data loss

📊 Downtime Characteristics

👉 Users typically don’t notice the migration

⚡ Challenges & How They’re Handled

High Memory Write Rate

• Problem: Pages keep changing rapidly
• Solution:
  • Multiple iterations
  • Adaptive throttling

Large Memory LPARs

• Problem: Longer copy time
• Solution:
  • Parallel memory transfer
  • Compression techniques

Network Constraints

• Migration speed depends on bandwidth
• High-speed links (10/25/100 GbE) improve performance

🧠 Key Insight

LPM maintains memory consistency by ensuring:

At the moment of switchover, the target system has an exact, up-to-date copy of all memory and CPU state.

🎯 Why This Matters

• Enables:
  • Zero/near-zero downtime maintenance
  • Load balancing across systems
  • Hardware upgrades without outages

🔑 Final Takeaway

Memory consistency in LPM is achieved through:

• Iterative copying
• Precise change tracking
• Final atomic synchronization