Oracle parallel recovery

Parallel recovery can speed up both instance recovery and media recovery. In parallel recovery, multiple parallel slave processes are used to perform recovery operations. The SMON background process reads the redo log files, and the parallel slave processes apply the changes to the datafiles.

In a serial recovery scenario, the SMON background process both reads the redo log files and applies the changes to the datafiles. This may take a considerably long time when multiple datafiles need to be recovered. However, when parallel recovery is being used, the SMON process is responsible only for reading the redo log files. The changes are applied to the datafiles by multiple parallel slave processes, thereby reducing the recovery time.

Recovery requires that the changes be applied to the datafiles in exactly the same order in which they occurred. This is achieved by single-threading the read phase of the recovery process by the SMON process. SMON reads the redo log files and serializes the changes before dispatching them to the parallel slave processes. The parallel slave processes then apply those changes to the datafiles in the proper order. Therefore, the reading of the redo log files is performed serially even during a parallel recovery operation.

读redo log 肯定是lgwr 的时,而smon 只是进行分发,我想肯定要file_id,block_id进行hash 的,否则cache buffer chains 的latch 竞争一定会非常严重,影响parallel的效率!

Trackback URL: http://blog.itpub.net//trackback.php?id=463506