the change seems to be introduced for extent alignment reasons (as already pointed out by Christo), and the behaviour can be controlled via event 60060 as I’ve recently learned, which disables or enables the extent alignment on allocation unit boundaries.

Enabling the event will revert to the pre-11g result for your test case, but has some more significant meaning for extent allocation in general (will publish an article soon).

Randolf

That sounds like a reasonable hypothesis, especially when you start thinking about ASM and AUs of 1MB and 4MB, and the impact of Exadata striping across cells. You could probably imagine seeing some sort of performance side effect if the extent boundaries on your objects were 64KB out of step with the AU boundary.

I think it’s more likely they did this for alignment reasons. I wonder if they also align segments on the MB boundary (when using auto-sizing) inside the datafiles as well.