I was talking about pinning workers, not fibers. Empirically, it is often better for performance to allow the OS to manage the hardware threads. Sometimes its w

I was talking about pinning workers, not fibers. Empirically, it is often better for performance to allow the OS to manage the hardware threads. Sometimes its worth pinning to a NUMA node, sometimes to a core, rarely to a cpu. Just reporting our results for many thousands of hours across a wide spectrum of test environments. I can understand it being counterintuitive, believe me it has been a shock. But it's true.