1. 8

  2. 2

    8 arenas per core seems like a lot. Even if every single CPU is calling into malloc, you only need ncpu arenas. How often are they expecting a CPU to get into malloc, then context switch away to another thread that then immediately calls malloc again? Times 8?