There are several types of checkpoints:

* Instance Recovery Checkpoint
* Media Recovery Checkpoint
* Thread Checkpoint
* Interval Checkpoint
* Tablespace Checkpoint
* Close Database Checkpoint
* Incremental Checkpoint
* Local Database Checkpoint
* Global Database Checkpoint

Of these, incremental checkpoints are the only type that only write to the controlfile and not to the datafile headers. All other types of checkpoint (including interval checkpoints if you still have them) write both to the controlfile and to the datafile headers. Rama Velpuri's book Oracle8i Backup & Recovery has a fairly complete description of the different types of checkpoints.

There are also cross-instance calls to checkpoint an object or block range, however, these are not true checkpoints in the sense that no checkpoint record is written anywhere. These are merely used to flush dirty buffers to disk prior to a direct read operation.

Does oracle perform a heavyweight checkpoint (update controlfile and datafile headers) at a log switch even if incremental checkpointing (FAST_START_IO_TARGET) is configured? I think that an SQL induced checkpoint (ALTER SYSTEM CHECKPOINT) causes a heavyweight checkpoint!