Originally posted by slimdave
I disagree on the composite partitioning -- maybe you are thinking of hash clusters, which would group the values.
Well i said of composite partitioning for his query as to "is there some way of using 2 levels of partitioning" and not my suggestion ..
They don't necessarily have to be -- the point is to get them clustered, so if on average you have 30 records per customer per partition you might fit them into one or two Oracle blocks instead of 30 Oracle blocks. Even if each Oracle block were split over 2 file system blocks you have still reduced the i/o's (on average) from the table by a factor of somewhere between 10 and 30
Well what if the smallest OS Blocks (.5K), which would mean 16 OS Blocks for a 8K Oracle Block, be split across N Disks?
Then do you still find any i/o reduction? -- Not neccessarly.
For this I would see no gain in ordering data by any field unless its a ordered structure.
"I Dont Want To Follow A Path, I would Rather Go Where There Is No Path And Leave A Trail."
"Ego is the worst thing many have, try to overcome it & you will be the best, if not good, person on this earth"