Databases Reference
In-Depth Information
performs a checkpoint to bring the datafile contents in line with the redo that
was written out for the committed transactions.
• If Oracle needs to read blocks requested by users into the cache and there is no
free space in the buffer cache, the blocks written out are the least recently used
blocks. Writing blocks in this order minimizes the performance impact of los‐
ing them from the buffer cache.
Log Writer (LGWR)
Writes the redo information from the log buffer in the SGA to all copies of the
current redo logfile on disk. As transactions proceed, the associated redo informa‐
tion is stored in the redo log buffer in the SGA. When a transaction is committed,
Oracle makes the redo information permanent by invoking the Log Writer to write
it to disk.
System Monitor (SMON)
Maintains overall health and safety for an Oracle instance. SMON performs crash
recovery when the instance is started after a failure and coordinates and performs
recovery for a failed instance when you have more than one instance accessing the
same database, as with Real Application Clusters. SMON also cleans up adjacent
pieces of free space in the datafiles by merging them into one piece and gets rid of
space used for sorting rows when that space is no longer needed.
Process Monitor (PMON)
Watches over the user processes that access the database. If a user process terminates
abnormally, PMON is responsible for cleaning up any of the resources left behind
(such as memory) and for releasing any locks held by the failed process.
Archiver (ARCn)
Reads the redo logfiles once Oracle has filled them and writes a copy of the used
redo logfiles to the specified archive log destination(s).
Up to 10 Archiver processes are possible—hence the notation ARC
n
. LGWR will
start additional Archivers as needed, based on the load, up to a limit specified by
the initialization parameter LOG_ARCHIVE_MAX_PROCESSES. By default, this
initialization parameter has a value of 2 and is rarely changed.
Checkpoint (CKPT)
Updates datafile headers whenever a checkpoint is performed.
Recover (RECO)
Automatically cleans up failed or suspended distributed transactions.
Dispatcher
Optional background processes used when shared server configurations are
deployed.
