Concurrent Execution

Doing work in parallel, or several things at once. There are three concurrency stories to know about, and they answer different questions:

• Auto-parallelization - pperl’s runtime detects parallelisable patterns in ordinary Perl and spreads them across cores without any user-level threading. This is the preferred answer for CPU-bound loops over large data.

• Coroutines and event loops - Coro, EV, Event and AnyEvent: many tasks that mostly wait - on sockets, timers, child processes - each written as straight-line code, taking turns in one interpreter. The answer for I/O-bound concurrency.

• Classical ithreads (not implemented) - Perl 5’s threads and threads::shared model. pperl deliberately does not provide it: use threads fails at compile time. The chapter teaches the model so that you can read existing code, and most programs that used ithreads are better rewritten against the auto-parallel path.

Start with the decision chapter if you have an existing threaded program and are deciding how to port it; start with parallel if you are writing new code and want to understand when pperl speeds your loop up, and with coroutines if your program spends its time waiting.

• Parallel Execution
  • How It Works
  • What Gets Parallelized
  • CLI Control
  • Performance
  • Limitations
  • How Reduction Detection Works
• Coroutines and event loops
  • When to use them
  • Starting and switching
  • Waiting for each other
  • Blocking on I/O with an event loop
  • Event loops
  • What each coroutine has of its own
  • C stacks
  • Porting from ithreads
  • See also
• Threads
  • pperl status - read this first
  • How this chapter is organised
  • Reference cross-links