Comment by nirushiv

I haven’t read through the code in detail but I can tell you “sub-nanosecond overhead” is misleading and marketing fluff. On first look, the measure seems to be some convoluted “time per thing” where the number of threads is far far smaller than the number of “thing”s