On 9t3 (maybe other versions too...)
I have carefully checked memory allocation strategy during an indigo render session for a rather simple scene (a transparent 32 multi faceted cylinder + sky only), and was a bit astonished by the way it goes.
I have noticed allocated memory ranging from 80 Mbytes to 120 Mbytes. Rough description : It goes from min to max in a few seconds stay at the max during say 30 seconds, then goes down to the min for the same period of time.
Why not forbid memory deallocation ? It would improve render time.... I am aware it must not be the same data that is allocated each time, but a memory pool manager could be used in order to avoid new/malloc, etc...
I'm aware memory allocation strategy is at the core of the algorithm, but for future versions, this could be the way to go...
My 2 cts worth
