From: Recurrent discussions

Some of the features which distinguish different approaches to
threading are listed below:

- Number of *concurrent* flows of control: generally, threads may
  potentially make use of multiple processors in order to allow
  several to execute concurrently.  That is, the model usually takes
  into consideration the possibility that there may be more than one
  flow of control active at any time.

- Scheduling policy: a thread scheduler may be pre-emptive, in which
  case a thread is put to sleep either when it waits upon some
  resource or runs for the full duration of its time quantum, or
  non-pre-emptive, in which case individual threads continue to run
  until they relinquish the processor themselves (either through
  waiting on a resource or calling the analogue of a sleep()
  function).

Systems which are non-pre-emptive and may only ever have a single
active flow of control (regardless of the number of processors
available) are referred to as coroutine systems.  Coroutine
programming requires quite a different approach from threads-based
programming, as many of the synchronisation and resource-sharing
problems which occur in threaded environments need never trouble the
coroutines programmer.