Using truss/strace to figure out memory bandwidth issues sounds pretty unreliable. I would not have guessed there was much correlation between memory syscalls t

Using truss/strace to figure out memory bandwidth issues sounds pretty unreliable. I would not have guessed there was much correlation between memory syscalls that truss/strace can observe (mmap/munmap, brk), and the CPU load/stores that consume memory bandwidth.