We might consider the creation of a monitoring thread. It would use some statistical usage in the different execution contexts to decide to start/stop schedulers in the MT schedulers. For example how many fibers are being enqueued/processed and given a ratio decide to wakeup a scheduler (enqueue more than we process), or on the contrary decide to stop a spinning one (fibers are processed faster than we enqueue).
We might consider the creation of a monitoring thread. It would use some statistical usage in the different execution contexts to decide to start/stop schedulers in the MT schedulers. For example how many fibers are being enqueued/processed and given a ratio decide to wakeup a scheduler (enqueue more than we process), or on the contrary decide to stop a spinning one (fibers are processed faster than we enqueue).