Overhead of `Threads.@threads`

I’m curious about nested schedulers in general. So, I wonder how spin-based task pool management interacts with Julia’s cooperative multitasking. Did you look at scenarios like

@sync for ...
    @spawn vmap!(...)
end

?

Since Julia’s parallel runtime randomizes scheduling, I think an interesting statistics for this is the median (“typical case”) than the minimum (“lucky path”).

1 Like