Hello,
To answer your question, V8 definitely supports memory segments. I don't know about papers but in the code each segment is an instance of Heap (see heap.cc and heap.h in the source). The 2 young spaces are in another contiguous segment of 8Mb.
Each segment of V8 are one of this kind (all instances of Heap):
* NewSpace: New object allocations go here. Capacity is 8MB. It's divided into two contiguous semispaces, and periodically collected using Cheney's algorithm. Uses bump-allocation.
* OldSpaces: These allocate memory in 64-128KB chunks of 8KB pages. They come in the following flavors:
* MapSpace: Exclusively for Maps (similar to js::Shape), and Maps *always* go here.
* CodeSpace: Exclusively for Code.
* DataSpace: For flat strings and numbers; can't reference objects, except for maps.
* PointerSpace: For anything not a Map or Code that can reference other objects.
* CellSpace: Seems to store global properties or something.
* LOSpace: For large objects, where "large" means it needs more than ~8KB of storage.
I have no proof for hotspot (their code is too complex) but I'm sure they have it too.
For the other question (adding elements to OrderedCollection), Eliot will answer it in more details, but the main idea is that you have the heap chopped into segments of a fixed size (1Mb in Cog, and�64-128KB chunks in V8), therefore for object bigger than a segment, you need to handle specifically the object, i.e. having a segment of the size of the object. If you add 30 000 elements to an OrderedCollection, the object will still be under 1MB so it is ok, now if you a 3 million of elements to this OrderedCollection, you need to handle specifically, Therefore you have performance loss for object of more than 1Mb but there are not common so you will not notice it (you can hardly have thousands of objects bigger than 1Mb allocated and deallocated regularly, and you need this amount to notice the performance loss).
An interesting point in V8 is that the data and pointers are not in the same spaces. It would mean in Cog, if I understand correctly, that byte and word objects are not in the same space as other objects. I guess this simplifies the GC logic. I don't know if they did that in VW.