here - Stefan-Marr.de

3. Which Concepts for Concurrent and Parallel Progr. does a VM need to Support?
actor-based language directly to application programs. The main operations
are directly provided by the instruction set. The send instruction implements
asynchronous message sends to a specified actor, i. e., Erlang process. The
wait and wait_timeout instructions are used to wait for incoming messages
on the message queue. To facilitate Erlang’s pattern matching, messages are
only removed from the message queue when the explicit remove_message instruction
is used. However, new Erlang processes are spawned by a primitive.
For use-cases like the Mnesia database, 12 Erlang also provides a set of primitives
for mutable concurrent hash-tables (ets, Erlang term storage 13 ) that can
be shared between actors. While this introduces a restricted notion of shared
state, it is not part of the core primitives suggested for general use.
Perl The Perl 14 VM offers support for so-called threads. However, this terminology
is not in line with the commonly accepted notion of threads. Perl’s
threads are more directly comparable to OS processes as they provide a nonshared
memory programming model by default. However, the VM provides
primitives in the threads::shared module to enable the use of shared data
structures between these threads. Data structures can only be shared between
threads after they have been explicitly prepared for sharing with the share()
primitive. To use this restricted form of shared memory, the VM provides
the Queue data structure for channel-like communication, as well as a lock
primitive, a Semaphore class, and primitives for condition variables.
Python As discussed in Sec. 3.1.2.1, Python’s main programming model is
based on threads and locks with a VM implementation that is constrained
by global interpreter lock (GIL) semantics. Since the GIL hinders any form of
parallel execution of Python bytecodes, VM support for process-based parallelism
was added. The standard library provides the multiprocessing module
to offer an abstraction for parallel programming similar to Perl. Python’s
VM includes primitives for channels in the form of a message queue (Queue),
as well as primitives for semaphores on top of which for instance the Lock and
Condition classes are built. These can operate on shared memory that has to
be requested explicitly between communicating processes and is restricted
to certain data structures. Value and Array are the shareable data structure
primitives directly exposed by the library. Value is by default a synchronized
12 http://www.erlang.org/doc/apps/mnesia/
13 http://www.erlang.org/doc/man/ets.html
14 http://perldoc.perl.org/perl.html
52