Checkpointing

A mechanism used in database recovery to reduce the time required to recover from a system failure.

The Problem It Solves

During normal operation, a DBMS keeps a log of all transactions. If a crash occurs, the system must scan the entire log to redo/undo transactions — which can be very slow for large logs. Checkpointing solves this by periodically creating a "save point".

What Happens During a Checkpoint?

• Suspends all active transactions temporarily
• Flushes all modified (dirty) buffers from memory to disk
• Writes a record to the log
• Resumes transactions

Recovery Using Checkpoints

After a crash, the system only needs to look at the log from the last checkpoint onwards:

• Transactions committed after the last checkpoint → REDO
• Transactions not committed at crash time → UNDO
• Transactions committed before the last checkpoint → NOTHING (already safe on disk)

Types of Checkpointing

Type	Description
Sharp Checkpoint	All dirty pages flushed at once; simple but causes a brief pause
Fuzzy Checkpoint	Pages flushed gradually; no pause, used in modern systems

Key Benefit

Checkpointing limits the recovery window, making crash recovery much faster and more efficient.