Previous | Next | Trail Map | Writing Java Programs | Threads of Control


Thread Priority

Previously in this lesson, we've claimed that threads run concurrently. While conceptually this is true, in practice it isn't. Most computer configurations have a single CPU, so threads actually run one at a time in such a way as to simulate concurrency. The execution of multiple threads on a single CPU, in some order, is called scheduling. The Java runtime supports a very simple, deterministic scheduling algorithm known as fixed priority scheduling. This algorithm schedules threads based on their priority relative to other "Runnable" threads.

When a Java thread is created, it inherits its priority from the thread that created it. You can also modify a thread's priority at any time after its creation using the setPriority() method. Thread priorities range between MIN_PRIORITY and MAX_PRIORITY (constants defined in class Thread). At any given time, when multiple threads are ready to be executed, the runtime system chooses the "Runnable" thread with the highest priority for execution. Only when that thread stops, yields, or becomes "Not Runnable" for some reason, will a lower priority thread start executing. If there are two threads of the same priority waiting for the CPU, the scheduler chooses them in a round-robin fashion.

The Java runtime system's thread scheduling algorithm is also preemptive. If at any time a thread with a higher priority than all other "Runnable" threads becomes "Runnable", the runtime system chooses the new higher priority thread for execution. The new higher priority thread is said to preempt the other threads.

The Java runtime system's thread scheduling scheme can be summed up with this simple rule:


Rule: At any given time, the highest priority runnable thread is running.

The 400,000 Micron Thread Race

This Java source code implements an applet that animates a race between two "runner" threads with different priorities. When you click the mouse down over the applet, it starts the two runners. The top runner, labelled "1", has a priority of 1 (the lowest possible thread priority in the Java system). The second runner, labelled "2", has a priority of 2.

Try This: Click over the applet below to start the race.

This is the run() method for both runners.

public int tick = 1;
public void run() {
    while (tick < 400000) {
        tick++;
    }
} 
This run() method simply counts from 1 to 400,000. The instance variable tick is public because the applet uses this value to determine how far the runner has progressed (how long its line is).

In addition to the two runner threads, this applet also has a third thread that handles the drawing. The drawing thread's run() method contains an infinite loop; during each iteration of the loop it draws a line for each runner (whose length is computed from the runner's tick variable), and then sleeps for 10 milliseconds. The drawing thread has a thread priority of 3--higher than either runner. So, whenever the drawing thread wakes up after 10 milliseconds, it becomes the highest priority thread, preempting whichever runner is currently running and draws the lines. Thus you can see the lines inch their way across the page

As you can see, this is not a fair race because one runner has a higher priority than the other. Each time the drawing thread yields the CPU by going to sleep for 10 milliseconds, the scheduler chooses the highest priority runnable thread to run; in this case, it's always the runner labelled "2". Here is another version of the applet that implements a "fair race", that is, both of the runners have the same priority and they have an equal chance of being chosen to run.

Try this: Again, click down with the mouse to start the race.

In this race, each time the drawing thread yields the CPU by going to sleep, there are two Runnable threads of equal priority--the runners--waiting for the CPU; the scheduler must choose one of the threads to run. In this situation, the scheduler chooses the next thread to run in a round-robin fashion.

Selfish Threads

The Runner class used in the races above actually implements "socially-impaired" thread behaviour. Recall the run() method from the Runner class used in the races above:
public int tick = 1;
public void run() {
    while (tick < 400000) {
        tick++;
    }
} 
The while loop in the run() method is in a tight loop. That is to say, once the scheduler chooses a thread with this thread body for execution, the thread never voluntarily relinquishes control of the CPU--the thread continues to run until the while loop terminates naturally or until the thread is preempted by a higher priority thread.

In some situations, having "selfish" threads doesn't cause any problems because a higher priority thread preempts the selfish one (just as the drawing thread in the RaceApplet preempts the selfish runners). However, in other situations, threads with CPU-greedy run() methods, such as the Runner class, can take over the CPU and cause other threads to have to wait for a long time before getting a chance to run.

Time-Slicing

Some systems fight selfish thread behaviour with a strategy known as time-slicing. Time-slicing comes into play when there are multiple "Runnable" threads of equal priority and those threads are the highest priority threads competing for the CPU. For example, this stand-alone Java program (which is based on the RaceApplet above) creates two equal priority selfish threads that have the following run() method.
public void run() {
    while (tick < 400000) {
        tick++;
        if ((tick % 50000) == 0) {
            System.out.println("Thread #" + num + ", tick = " + tick);
        }
    }
}    
This run() contains a tight loop that increments the integer tick and every 50,000 ticks prints out the thread's identifier and its tick count.

When running this program on a time-sliced system, you will see messages from both threads intermingled with one another. Like this:

Thread #1, tick = 50000
Thread #0, tick = 50000
Thread #0, tick = 100000
Thread #1, tick = 100000
Thread #1, tick = 150000
Thread #1, tick = 200000
Thread #0, tick = 150000
Thread #0, tick = 200000
Thread #1, tick = 250000
Thread #0, tick = 250000
Thread #0, tick = 300000
Thread #1, tick = 300000
Thread #1, tick = 350000
Thread #0, tick = 350000
Thread #0, tick = 400000
Thread #1, tick = 400000
This is because a time-sliced system divides the CPU into time slots and iteratively gives each of the equal-and-highest priority threads a time slot in which to run. The time-sliced system will continue to iterate through the equal-and-highest priority threads allowing each one a bit of time to run until or more of them finish or until a higher priority preempts them. Notice that time-slicing makes no guarantees as to how often or in what order threads are scheduled to run.

When running this program on a non-time-sliced system, however, you will see messages from one thread finish printing before the other thread ever gets a chance to print one message. Like this:

Thread #0, tick = 50000
Thread #0, tick = 100000
Thread #0, tick = 150000
Thread #0, tick = 200000
Thread #0, tick = 250000
Thread #0, tick = 300000
Thread #0, tick = 350000
Thread #0, tick = 400000
Thread #1, tick = 50000
Thread #1, tick = 100000
Thread #1, tick = 150000
Thread #1, tick = 200000
Thread #1, tick = 250000
Thread #1, tick = 300000
Thread #1, tick = 350000
Thread #1, tick = 400000
This is because a non-time-sliced system chooses one of the equal-and-highest priority threads to run and allows that thread to run until it relinquishes the CPU (by sleeping, yielding, finishing its job) or until a higher priority preempts it.

Try this: Compile and run the RaceTest and SelfishRunner classes on your computer. Can you tell if you have a time-sliced system?

As you can imagine, writing CPU-intensive code can have negative repercussions on other threads running in the same process. In general, you should try to write "well-behaved" threads that voluntarily relinquish the CPU periodically and give other threads an opportunity to run. In particular, you should never write Java code that relies on time-sharing--this will practically guarantee that your program will give different results on a different computer system.

A thread can voluntarily yield the CPU (without going to sleep or some other drastic means) by calling the yield() method. The yield() method gives other threads of the same priority a chance to run. If there are no equal priority threads in the "Runnable" state, then the yield is ignored.

Try this: Rewrite the SelfishRunner class to be a PoliteRunner by calling the yield() method from the run() method. Be sure to modify the main program to create PoliteRunners instead of SelfishRunners. Compile and run the new classes on your computer. Now isn't that better?

Summary


Previous | Next | Trail Map | Writing Java Programs | Threads of Control