为什么buffer cache不起作用

Yong Huang

> It becomes worse situation when paralleled by application itself,...

If you manually split the work, why not consider partitioning the table? Then each parallel slave handles a few partitions only.

wolfop

liqy103 发表于 2012-8-24 15:18
It becomes worse situation when paralleled by application itself, e.g. we have 30 streams to run sam ...

That is a wrong usage to deal mass data in 11GR2. I have seen many this wrong way in production. Why not use parallel query in one query not partition by the application just like most MPP database.
Another simple way may help is to set paralle_degree_policy to AUTO, but you need to do some test and IO calibrate.  
Another point is this is not a 11G2R new feature, in 10GR2 there are also direct path reads. You shoud try to find why the optimizer choose direct path read.

liqy103

wolfop 发表于 2012-8-26 11:08
That is a wrong usage to deal mass data in 11GR2. I have seen many this wrong way in production. W ...

This is a vendor-provide application, as DBA we have little influence on code change to those "sound" vendors.

winsonjie

thanks for your sharing.