VirtualMemoryTracker currently allocate all of its linked list nodes into the C heap. It's faster to allocate them in an Arena, as you avoid the full malloc calls each time and traversing them becomes faster (fewer cache misses).