Well, I’m giving this talk related to Exadata at the Collaborate 13 usergroup conference on Monday. I’ve spent a lot of time – probably too much time – preparing the slides and practicing the talk. I first gave this talk a year ago at our office and then again last October at a conference in North Carolina. But, I’ve never been happy with it. I feel a lot better about it now because the slides and the associated notes have a lot of information on them. I got a lot of great feedback from several people and I’ve made changes accordingly.
But having spent too much time reviewing this I think there are a small number of important concepts that I’m really trying to get across:
- Exadata Smart Scans bypass the block buffer cache
- Exadata Smart Scans happen instead of FULL scans
- Make indexes invisible or increase optimizer_index_cost_adj to encourage FULL scans
- Exadata Smart Scans tend to be part of a HASH JOIN
- HASH JOINs can be sped up by adding PGA memory – pga_aggregate_target
- You can free memory by reducing the size of the block buffer cache – sga_max_size
So, this is the short version of my 45 minute talk More PGA, less SGA – my talk in four words!