Hi We do use PX. Constantly around 32 to 64 slaves are used (depends on the node). But not observing much sorting or hash join contention on the disks. Thanks -- LSC On Sun, Apr 18, 2010 at 11:55 PM, Greg Rahn <greg@xxxxxxxxxxxxxxxxxx> wrote: > So I take it you are not using PX on this database (doesn't seem that > big if you have just 6GB PGA) -- you are running stuff serial and > relying on the buffer cache for read performance? > > Given a large enough database and system with enough disk/channel > bandwidth it may make more sense to make large dimensions PARALLEL. > > On Sat, Apr 17, 2010 at 4:59 AM, LS Cheng <exriscer@xxxxxxxxx> wrote: > > I have 6GB PGA set. Before I check the system it was 3GB and had some > > sorting contention but with 6GB I barely observe large value for > multipass > > system statistics, most is memory and some one pass. > > > > The large shared pool is because the problems I have observed, probably a > > RAC bug. > > > > The large buffer cache is for the dimension tables, some of them has > million > > of rows and ncreasing :-S > > -- > Regards, > Greg Rahn > http://structureddata.org >