A checkpoint is the writing by the (database writer) process of all modified buffers in the buffer cache to the database . Data file headers are also updated with the latest checkpoint , even if the file had no changed blocks, as well as the control files.

Checkpoints occur AFTER (not during) every switch and also at intervals specified by initialization parameters.

Set parameter LOG_CHECKPOINTS_TO_ALERT=TRUE to observe checkpoint start and end times in the database alert log.

Checkpoints can be forced with the ALTER SYSTEM CHECKPOINT; command.

• A number of dirty database buffers covered by the log being checkpointed are written to the data files by DBWn. The number of buffers being written by DBWn is determined by the FAST_START_MTTR_TARGET parameter, if specified.  (Oracle official doc) .

**AskTom***:  a checkpoint is the act of flushing modified, cached database blocks to disk. Normally, when you make a change to a block -- the modifications of that block are made to a memory copy of the block. When you commit -- the block is not written (but the REDO LOG is -- that makes it so we can "replay" your transaction in the event of a failure). Eventually, the system will checkpoint your modified blocks to disk. there is no relationship between "checkpoint" and sid and instance recovery does not imply "checkpoint". a checkpoint reduces the amount of time it takes to perform instance recovery.......