How to calculate overall computation time for a multi-threaded process

Question

I have a set of tasks, let\'s call it T[], where each task T[i] needs a certain amount of time t(T[i]) to be processed. The tasks are bein

Answer 1

Simulating the test run is the solution when the execution order is (roughly) deterministic. I just took my real processing code, and replaced it with a simple Thread.sleep, while sleeping for the time the task is expected to take to process (just interpreted as milliseconds to scale it down). In the end, I just scaled up the time it took again and the result is quite good. I ran it with nearly 100 tasks with vastly different execution times, on 5 threads. It estimated 1 hour 39 minutes, and the real run was off only by 3 minutes.

            long startSim = currentTimeMillis();
            List taskTimes = parallelTests.getRuntimesForAllTests(); // ordered from longest time
            ThreadPoolExecutor simulationExecutor = (ThreadPoolExecutor) Executors.newFixedThreadPool(threadCount);
            taskTimes.forEach(taskTime -> simulationExecutor.submit(() -> {
                try {
                    Thread.sleep(taskTime); // this is really seconds, but we just take it as milliseconds
                } catch (InterruptedException e) {
                    e.printStackTrace();
                }
            }));
            simulationExecutor.shutdown();
            simulationExecutor.awaitTermination(1, MINUTES);
            long stopSim = currentTimeMillis();
            long timeNeeded = stopSim - startSim;
            // now just multiply it *1000 to scale it up to seconds again, and that's your result