Exploiting Multi-core CPUs
Today's computers typically have multi-core CPUs. This means that they are able to execute multiple streams of instructions simultaneously.
We can take advantage of multiple CPU cores by creating threads. A thread is a stream of instructions executed within a program. One way to think of a thread is that it is a "virtual CPU". When a program has multiple threads, it has multiple virtual CPUs executing parts of the program at the same time.
Runnable
The Runnable interface describes a computational task. It is defined as follows:
public interface Runnable {
public void run();
}When your program has a large computation to split into parts, you should define your own "computation task" class which implements the Runnable interface. Your program will create one instance of your task class for each "chunk" of the computation.
Thread
The Thread class represents a thread (virtual CPU). A thread takes a reference to an object implementing the Runnable interface, and executes its run method in a separate thread.
There are three important Thread methods that you will use.
The Thread class's constructor takes a reference to the Runnable object whose run method you want the new thread to execute. Note that creating the Thread object does not actually start the thread.
The Thread class's start method actually starts a new thread. The new thread will execute the run method of the Runnable object passed to the Thread's constructor.
The Thread class's join method can be called to wait for the thread to complete. In other words, the join method returns after the Runnable object's run method returns.
Note that the join method can throw a checked exception called InterruptedException.
Fork/Join Parallelism
A very common pattern for parallel computation is fork/join parallelism. In programs that employ this form of computation, the main program creates some number of parallel "worker" threads, starts them, and then waits for all of them to complete.
Here is how fork/join parallelism can be implemented in Java. This code example assumes that there is a class called ComputeTask that implements the Runnable interface and carries out a given chunk of the computation. This example creates 4 threads, allowing parallel computation on up to 4 CPU cores simultaneously.
// Split the computation into 4 tasks
ComputeTask[] tasks = new ComputeTask[4];
for (int i = 0; i < 4; i++) {
tasks[i] = new ComputeTask(...data that describes the task...);
}
// create the threads that will execute the tasks,
// and start each thread
Thread[] threads = new Thread[4];
for (int i = 0; i < 4; i++) {
threads[i] = new Thread(tasks[i]);
threads[i].start();
}
// wait for all of the threads to complete
try {
for (int i = 0; i < 4; i++) {
threads[i].join();
}
} catch (InterruptedException e) {
// this should not happen
System.err.println("A thread was interrupted");
}
// Now, we can look at the results of each compute task,
// and combine them into a single overall resultCommunication Between Threads
In fork/join parallelism, the computation threads are completely independent of each other. This is the ideal case for a parallel computation: each parallel task does not need to communicate with any other parallel task. Such parallel computations tend to be easy to implement, and execute very efficiently.
Some parallel computations, however, require communication between computation tasks. Threads executing in parallel can communicate with each other using a shared data structure. For example, a queue (list) could be used to communicate values from one thread to the next.
When a data structure is shared between threads, some form of synchronization must be used to ensure that the threads are not accessing and modifying the data structure at the same time. We will not be investigating thread synchronization in this course, but the basic idea is that "locks" can be used to temporarily give one thread exclusive access to the shared data structure.
(If you are interested in learning more, do a web search for "mutex", "critical section", and "java synchronization".)
