from 1.93:
* Following is a brief list of features that distinguish this malloc
* implementation:
*
* + Multiple arenas are used if there are multiple CPUs, which reduces lock
* contention and cache sloshing.
*
* + Cache line sharing between arenas is avoided for internal data
* structures.
*
* + Memory is managed in chunks, rather than as individual pages.
*
* + Allocations are region-based; internal region size is a discrete
* multiple of a quantum that is appropriate for alignment constraints.
* This applies to allocations that are up to half the chunk size.
*
* + Coalescence of regions is delayed in order to reduce overhead and
* fragmentation.
*
* + realloc() always moves data, in order to reduce fragmentation.
*
* + Red-black trees are used to sort large regions.
*
* + Data structures for huge allocations are stored separately from
* allocations, which reduces thrashing during low memory conditions.
*