Python threads: backed by OS threads?

[David Arnold]

-->"Courageous" == Courageous  <jkraska1 at san.rr.com> writes:

  Courageous> I just read that. It suggests that the threads may be
  Courageous> mutually locking eachother out some of the time, but it
  Courageous> does not say that they will not see a performance gain
  Courageous> once you discount the contention issues.

whenever the threads are within the python interpreter itself, the are
mutually exclusive.

calls out to C extensions tend to release the interpreter lock, unless
the operation is purely computational and short-lived.  so all I/O,
time.sleep(), etc release the lock.

the end result is that an I/O bound process will see substantial
improvements by threading.  CPU-bound processes will not, regardless
of multi-processor machines.

there is ongoing work to support so-called "free-threading" of the
interpreter, usually discussed within the Thread-SIG.  patches were
available for 1.4, and the underpinnings of the interpreter have been
gradually gaining support for it.  the rough plan is to have it
available as a compile-time option in the not-too-distant future.



d