what is the reason of design decision of bypassing the buffer cache for parallel query?

Dear Gurus ,

as far as i know and tested , when i issue a parallel select statement , oracle ,instead of looking at database buffer cache first, it directly requests the data from the disk system. And when it reads the data , instead of putting it into db buffer cache ,it copies it into the process pga. So there is no db buffer cache benefit if the same parallel query runs several times since db buffer cache does not contain any data related to this table.

If the above statement is false , pls correct me. All comments are wellcomed.

In this situation , if the underlying file system uses file caching such as sun ufs ,or linux ext3 , it is possible that the data that is requested from oracle can be retrieved by using the file cache.

I use ocfs which uses direct io and bypasses this buffer cache, so i neither use oracle cache nor file cache but directly read data from the emc disk array . ( Only the emc disk cache exists.)

Isnt it better to have a cache at one of these levels? Why did oracle design parallel query like this? Why does not parallel query cache data?

Kind Regards,
tolga Received on Wed Jun 01 2005 - 01:50:54 CDT