An experiment in simulating parallel systems perfectly. Provides a green threading model and a controlled clock allowing for parallel execution of arbitrarily many threads. Virtual time advances as fast as the system can go, which allows you to simulate long-running or very large-scale systems quickly, without error.
If a thread says it wants to wake up at time t, it awakes at exactly time t, is guaranteed to see all memory side effects from all threads which ran at earlier times, and that only threads scheduled to awaken at t are currently running. The clock only advances when threads choose to sleep, allowing you precise control over the simulated evolution of the system. Naturally, one can trivially recover stochastic ordering dynamics if desired.
(use 'timelike.scheduler) (def counter (atom 0)) (dotimes [i 100] (thread ; All these swaps run concurrently at time 0. (swap! counter inc) (time) ; returns 0 ; Blocks for a random amount of time (sleep (rand)) ; The scheduler guarantees that this part of the code executes ; one at a time (unless two rands were the same) (time) ; might return 0.0023453, or 0.99234. (swap! counter inc) ; And now let's synchronize: all threads wake up simultaneously at 100 seconds: (sleep-until 100) (time) ; returns 100 (swap! counter inc))) ; Blocks until all threads have completed. This should take only a few milliseconds. (await-completion) (prn @counter) ; Prints 300: 100 threads performing 3 increments each.
Copyright © 2013 Kyle Kingsbury
Distributed under the Eclipse Public License, the same as Clojure.