Distributed Computing: Distributed System Checkpoints and it’s type-checkpoint levels its tools

Distributed Computing: Distributed System Checkpoints and it’s type-checkpoint levels its tools.

Distributed Computing: Checkpoints in Distributed Systems

 What is Checkpointing in Distributed Systems?

Checkpointing is a fault-tolerance mechanism in distributed computing that periodically saves the system state. If a failure occurs, the system can restart from the last checkpoint instead of starting from scratch.

Key Idea:
 Saves system state at intervals.
 Reduces computation loss during failures.
 Speeds up recovery in distributed systems.

 Types of Checkpoints in Distributed Systems

Coordinated Checkpointing

Definition: All nodes in the system synchronize and save their states together.

Ensures consistency (no orphan or lost messages).
 Used in global snapshots.
Slower due to coordination overhead.

Example:

• Two-Phase Commit (2PC)
• Chandy-Lamport Algorithm

 Uncoordinated Checkpointing

Definition: Each process takes checkpoints independently without synchronization.

Faster, no coordination required.
Risk of cascading rollbacks (domino effect).

Example:

• Individual process backups.

 Communication-Induced Checkpointing

Definition: A hybrid approach where checkpoints are triggered based on message passing.

Prevents inconsistent states.
Avoids the domino effect from uncoordinated checkpointing.
Higher overhead due to message tracking.

Example:

• Log-based checkpointing in distributed databases.

 Application-Level Checkpointing

Definition: Checkpoints are managed at the software level rather than system level.

 Allows customized checkpointing in applications.
 Efficient for high-performance computing (HPC).
Requires developer implementation.

Example:

• MPI (Message Passing Interface) checkpointing.

 Checkpoint Levels in Distributed Systems

Process-Level Checkpointing: Saves state of individual processes.
System-Level Checkpointing: Saves the entire OS state.
Application-Level Checkpointing: Saves the state at the application level.

 Tools for Checkpointing in Distributed Systems

1. DMTCP (Distributed MultiThreaded Checkpointing)

• Application-level checkpointing.
• Supports parallel computing systems.

2. CRIU (Checkpoint/Restore in Userspace)

• Process-level checkpointing for Linux.
• Saves process state & resumes execution.

3. BLCR (Berkeley Lab Checkpoint/Restart)

• Kernel-level checkpointing for HPC systems.
• Works with MPI applications.

4. Hadoop Checkpointing

• Used in HDFS (Hadoop Distributed File System) for fault tolerance.

 Conclusion

Checkpointing reduces system failures’ impact by allowing recovery from saved states. Different types & tools are used based on the system requirements (speed, reliability, and overhead).

Would you like code examples or real-world use cases?