Solution for Programming Exercise 12.1
This page contains a sample solution to one of the exercises from Introduction to Programming Using Java.
Exercise 12.1:
Subsection 12.1.3 discusses the need for synchronization in multithreaded programs, and it defines a ThreadSafeCounter class with the necessary synchronization. Is this really important? Can you really get errors by using an unsynchronized counter with multiple threads? Write a program to find out. Use the following unsynchronized counter class, which you can include as a nested class in your program:
static class Counter {
int count;
void inc() {
count = count+1;
}
int getCount() {
return count;
}
}
Write a thread class that will call the inc() method in this class a specified number of times. Create several threads, start them all, and wait for all the threads to terminate. Print the final value of the counter, and see whether it is correct.
Let the user enter the number of threads and the number of times that each thread will increment the counter. You might need a fairly large number of increments to see an error. And of course there can never be any error if you use just one thread. Your program can use join() to wait for a thread to terminate (see Subsection 12.1.2).
The necessary thread class is very simple to write. In my program, it's a static nested class that uses two global variables, which represent the counter and the number of times that the thread is to increment the counter:
static Counter counter; // The counter that will be incremented.
static int numberOfIncrements; // Number of times each thread will increment it.
static class IncrementerThread extends Thread {
public void run() {
for (int i = 0; i < numberOfIncrements; i++) {
counter.inc();
}
}
}
Note that since the counter is a global static variable, it is used by all threads of type IncrementerThread. It is a shared resource.
In the main() routine, after getting the number of threads and the number of increments from the user, we need to create the threads, start them, and wait for them to terminate. We have to do the part about starting all the threads before we get to the part about waiting for threads to terminate, so we need an array to hold the threads:
IncrementerThread[] workers = new IncrementerThread[numberOfThreads];
counter = new Counter();
for (int i = 0; i < numberOfThreads; i++)
workers[i] = new IncrementerThread();
for (int i = 0; i < numberOfThreads; i++)
workers[i].start();
/* Wait for all threads to terminate. */
for (int i = 0; i < numberOfThreads; i++) {
try {
workers[i].join();
}
catch (InterruptedException e) {
}
}
(I hope that it's clear that you can't start a thread, wait for it to finish, start another thread, wait for it to finish, and so on. If you did that, you'd only have one thread running at any given time, and you would certainly not see any errors in the counter!)
My program actually runs in a loop, so the user can run several experiments without restarting the program every time.
In my own experiments, I saw a lot of errors when the number of increments per thread was one million, even with just two threads. For small numbers of increments, less than about 1000, I never saw any errors. (This might be because the threads didn't run long enough to actually get more than one running at a time.) I saw a few errors when I used a large number of threads with 1000 increments per thread.
Making inc() into a synchronized method completely solves the problem. If you try that, you should find that the program takes noticeably longer to run, because of the extra overhead associated with the synchronization. Simply declaring the variable count to be volatile does not fix the problem, since the race condition still exists. (In fact, curiously, I found that the number of errors increased when I made count volatile, possibly because accessing the volatile variable can cause a thread to block briefly and give another thread the chance to jump in.)
This exercise shows that synchronization errors can happen. In many programs, they would be very rare, but serious when they occur. Unfortunately, when trying to track down a bug that is due to a synchronization error, it might be very difficult to reproduce the error, since such errors depend on accidents of timing between two threads that are running independently. This makes it all the more important to analyze your program carefully and get it right when you are writing it in the first place, rather than count on your ability to debug it later.
import java.util.Scanner;
/**
* A demo program to see how hard it is to get an
* error while using an unsynchronized counter with
* several threads. The program runs one or more
* threads. Each thread increments the counter a
* specified number of times. After all the threads
* have completed, the value of the counter is
* printed out so it can be compared to the correct
* value. The user specifies the number of threads
* and the number of times each thread increments the
* counter. (To see an error, the number of increments
* probably has to be pretty large. Try 1000000.)
*/
public class UnsynchronizedCounterTest {
/**
* A class representing a counter with a method for
* incrementing the counter. No synchronization is
* used, so this counter is not "thread-safe".
*/
static class Counter {
int count;
void inc() {
count = count+1;
}
int getCount() {
return count;
}
}
static Counter counter; // The counter that will be incremented.
static int numberOfIncrements; // Number of times each thread will increment it.
/**
* The class that defines one of the threads. The thread
* simply increments counter numberOfIncrements times, in
* a for loop.
*/
static class IncrementerThread extends Thread {
public void run() {
for (int i = 0; i < numberOfIncrements; i++) {
counter.inc();
}
}
}
/**
* The main program runs in a loop until the user wants to exit.
* Each time through the loop, it runs one experiment. It gets
* the number of threads and the number of increments per thread
* from the user. It creates and starts the threads, and then
* waits for them all to finish. It prints the final value of
* the counter, as well as the expected value. The program ends
* when the user enters a number less than or equal to zero as
* the number of threads.
*/
public static void main(String[] args) {
Scanner in = new Scanner(System.in); // For reading the user's inputs.
while (true) {
/* Get number of threads and number of increments per thread
* from the user. Exit if number of threads is <= 0. */
System.out.println();
System.out.print("How many threads do you want to run (Enter 0 to end)? ");
int numberOfThreads = in.nextInt();
if (numberOfThreads <= 0)
break;
do {
System.out.println();
System.out.println("How many times should each thread increment the counter? ");
numberOfIncrements = in.nextInt();
if (numberOfIncrements < 1) {
System.out.println("Number of increments must be positive.");
numberOfIncrements = 1;
}
} while (numberOfIncrements <= 0);
System.out.println();
System.out.println("Using " + numberOfThreads + " threads.");
System.out.println("Each thread increments the counter "
+ numberOfIncrements + " times.");
/* Create the threads and start them. */
System.out.println();
System.out.println("Working...");
System.out.println();
IncrementerThread[] workers = new IncrementerThread[numberOfThreads];
counter = new Counter();
for (int i = 0; i < numberOfThreads; i++)
workers[i] = new IncrementerThread();
for (int i = 0; i < numberOfThreads; i++)
workers[i].start();
/* Wait for all threads to terminate. */
for (int i = 0; i < numberOfThreads; i++) {
try {
workers[i].join();
}
catch (InterruptedException e) {
}
}
/* Display the results. */
System.out.println("The final value of the counter should be "
+ (numberOfIncrements*numberOfThreads));
System.out.println("Actual final value of counter is: " + counter.getCount());
System.out.println();
System.out.println();
} // end while
} // end main()
} // end class UnsynchronizedCounterTest