Using KernelShark to analyze the real-time scheduler

Interesting. Looking at the second table of 8 tasks it's clear there are 4 CPUs. The 4 highest priority threads have much fewer than 1000 migrations -- 150 or less -- while the remaining threads have 1000 or more. The number of non-voluntary pre-emptions is similarly distributed (presumably because those are involved in the push/pull migrations).