Faster Virtual Method Dispatch

[Craig Black]

> I'm not entirely familiar with the concept of allocators as you describe. 
> Is this similar to memory pooling?  I always skipped over that portion of 
> the C++ STL; way too much complexity for my small brain =P.

Instead of calling malloc for each allocation unit, you could write an 
allocator class that would allocate memory in large chunks.  Then when you 
want an allocation unit, you request it from the allocator.  The allocator 
simply hands out the memory, an marks it as used.

The hierarchical memory system you describe lends itself to using such 
custom allocators, because you would never have to call free for each unit. 
You could call free at the allocator level instead.  In this way, you would 
only call malloc and free for large chunks of memory.

You could also inform each allocator how much memory to reserve when they 
are created, since you may have a good idea of how much memory a given task 
will require.

Did that help?

-Craig