observations: 1) we’re cpu bound on a single core during all of the super slow ops 2) according to

time

, we’re doing 6x the syscall overhead