By default, start a number of workers equal to the number of CPUs. (#430)

* By default, start a number of workers equal to the number of CPUs.

* Fix stress tests.

doc/source/api.rst

@@ -35,6 +35,12 @@ If there are GPUs available on the machine, you should specify this with the
 By default, Ray will use ``psutil.cpu_count()`` to determine the number of CPUs,
 and by default the number of GPUs will be zero.
 
+Instead of thinking about the number of "worker" processes on each node, we
+prefer to think in terms of the quantities of CPU and GPU resources on each
+node and to provide the illusion of an infinite pool of workers. Tasks will be
+assigned to workers based on the availability of resources so as to avoid
+contention and not based on the number of available worker processes.
+
 Connecting to an existing cluster
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 

doc/source/serialization.rst

@@ -87,7 +87,7 @@ This can be addressed by calling ``ray.register_class(Foo)``.
 
   import ray
 
-  ray.init(num_workers=10)
+  ray.init()
 
   # Define a custom class.
   class Foo(object):