Thanks Yechiel. That's my gut feel as well.The only minor snag I can see is that AFAIK the actual I/O size used by RMAN is not documented anywhere. If it is doing 512 units in scatter/gather I/O, then it's virtually the same as 8K blocks - assuming the db is 8k.
Given that it is mostly doing big sequential as you note, it shouldn't matter at all if the last block gets half-filled in case it's doing 8k block I/O.
Ah, I wish I had in Unix the ability to trace actual raw I/O sizes no matter what, like I could in the mainframes...
-- Cheers Nuno Souto in sunny Sydney, Australia dbvision@xxxxxxxxxxxx Yechiel Adar wrote,on my timestamp of 28/03/2011 5:52 PM:
No experience, just a thought. It is used for big sequential writes and no random reads or writes. In this case I will used the biggest block I can to minimize the I/Os to the disk. The reason we use small blocks is to minimize the redundant data transferred to memory when you need only one record in the block. When you write or read whole blocks, in big sets, use the block size that will minimize I/O.
-- //www.freelists.org/webpage/oracle-l