Download presentation
Presentation is loading. Please wait.
1
1 Java Threads Instructor: Mainak Chaudhuri mainakc@cse.iitk.ac.in
2
2 Java threads Two ways to create a thread –Extend the Thread class and override the public run method –Implement a runnable interface with a public run method class MyThread extends Thread { private int tid; private String name; private long startTime; public MyThread (String name, int tid) { super (name); this.tid = tid; this.name = new String(name); startTime = System.currentTimeMillis(); } // next slide
3
3 Java threads public void run () { // override System.out.println (“ Hi! This is ” + getName() + “.”); } public long age () { return (System.currentTimeMillis () – startTime); } } // end class class MyFirstThreadDemo { public static void main (String arg[]) { Integer n = new Integer (arg[0].trim()); int numThreads = n.intValue(); // next slide
4
4 Java threads int i; MyThread t[] = new MyThread[numThreads]; for (i=0; i<numThreads; i++) { t[i] = new MyThread (“Thread ”+i, i); t[i].start(); } } // end main } // end class
5
5 Java threads via interface class MyThread implements Runnable { private int tid; private String name; private long startTime; public MyThread (String name, int tid) { this.tid = tid; this.name = new String(name); startTime = System.currentTimeMillis(); } public void run () { System.out.println (“ Hi! This is ” + Thread.currentThread().getName() + “.”); } //next slide
6
6 Java threads via interface public long age () { return (System.currentTimeMillis () – startTime); } } // end class class MyFirstThreadDemo { public static void main (String arg[]) { Integer n = new Integer (arg[0].trim()); int numThreads = n.intValue(); int i; // next slide
7
7 Java threads via interface MyThread mt[] = new MyThread[numThreads]; Thread t[] = new Thread[numThreads]; for (i=0; i<numThreads; i++) { mt[i] = new MyThread (“Thread ”+i, i); t[i] = new Thread (mt[i], “Thread ”+i); t[i].start(); } } // end main } // end class
8
8 Announcements Final exam on 19 th November 1215 to 1515 Seating arrangement: Y4, Y5, Y7001 to Y7085L1 OROS Y7092 to Y7233L2 ERES Y7234 to Y7388L16 OROS Y7391 to Y7518L17 ERES Please take seats 15 minutes before the scheduled time
9
9 Java thread states Java threads have typically seven states –New: just after a thread is created –Runnable: after the start() method is invoked. The thread is placed in the runnable set or the ready queue. When scheduled, the run() method will be executed. –Running: executing the code in the run() method. May call its yield() method to put itself back in the runnable set. –Suspended: after the suspend() method is called. May be called by itself or by some other thread. It is placed back on runnable set only if some other thread calls its resume() method.
10
10 Java thread states Seven states (continued) –Blocked: after the sleep() method or the wait() method or the join() method is called to join with a thread (at a barrier) that is yet to arrive at the join point or when it does blocking I/O (e.g., reading from keyboard). The thread transitions back to runnable state when the blocking call is over. –Suspended-blocked: If a blocked thread is suspended by some other thread. It returns back to suspended or blocked state depending on whether the blocking finishes before a call to resume() or not. –Dead: On completion of the run() method or a call to stop() method.
11
11 Java thread scheduling Every Java thread gets a priority –Inherits the same priority from the creating thread. In our example, all threads have same priority as the “main thread”. –Priorities range from MIN_PRIORITY to MAX_PRIORITY defined in Thread class. The default priority is NORM_PRIORITY. –Use setPriority and getPriority methods to change and query a thread’s priority. –Among all the threads in the runnable set, the highest priority thread is allowed to run until it blocks, yields, gets suspended, or a new thread of higher priority enters the runnable set. In the last case the running thread goes back to runnable set
12
12 Java thread scheduling Some implementations carry out a round- robin scheduling (e.g., JDK 1.1 for Windows 95/NT) –Every thread gets a time slice to execute –Once the time slice of the thread expires, it is put back in the runnable set and the next thread is given a chance (higher priority threads are considered first) –If a thread blocks, yields, or gets suspended before the time slice expires, the next thread is scheduled for execution
13
13 More example class MyThread extends Thread { private int tid; private String name; private long startTime; private long screamingInterval; public MyThread (String name, int tid, long screamingInterval) { super (name); this.tid = tid; this.name = new String(name); startTime = System.currentTimeMillis(); this.screamingInterval = screamingInterval; } // next slide
14
14 More example public void run () { // override long count = 0; while (true) { if (count % screamingInterval == 0) { System.out.println (“ Hi! This is ” + getName() + “.”); } count++; } public long age () { return (System.currentTimeMillis () – startTime); } } // end class
15
15 More example class MyThreadDemo { public static void main (String arg[]) { Integer n = new Integer (arg[0].trim()); int numThreads = n.intValue(); int i; MyThread t[] = new MyThread[numThreads]; for (i=0; i<numThreads; i++) { t[i] = new MyThread (“Thread ”+i, i, (i+1)*1000000); //t[i].setPriority (t[i].getPriority()+i); t[i].start(); } } // end main } // end class
16
16 Typical output snapshot Hi! This is Thread 0. Hi! This is Thread 1. Hi! This is Thread 0. Hi! This is Thread 1. Hi! This is Thread 0. Hi! This is Thread 1. Hi! This is Thread 0. Hi! This is Thread 1. Hi! This is Thread 0.
17
17 Physical resources Parallelism ultimately depends on availability of multiple physical CPUs –You get true concurrency with “multiprocessors” –With single CPU, you get time-shared concurrency (at any point in time only one thread can run) –Today small-scale multiprocessors are commodity –Chip-multiprocessors or so called multi-core processors come with two, four, or eight processors on a single chip (soon 16 processors)
18
18 Data race How to share a variable? –Create a single class containing the variables that you want to share –Create one instance of this object and attach multiple threads to it –Let us see what happens if we try to update a shared variable concurrently in multiple threads –This is known as a data race as multiple threads may have a race trying to update the same data
19
19 Data race class MyThread implements Runnable { private long sum; public MyThread () { sum=0; } public void run () { int i; for (i=0; i<100000; i++) { sum++; } System.out.println(" : Sum=" + sum); } // next slide
20
20 Data race public long GetSum () { return sum; } } // end class class MyDataRaceDemo { public static void main (String arg[]) throws java.lang.InterruptedException { Integer n = new Integer (arg[0].trim()); int numThreads = n.intValue(); int i; // next slide
21
21 Data race MyThread mt = new MyThread(); Thread t[] = new Thread[numThreads]; for (i=0; i<numThreads; i++) { t[i] = new Thread (mt, "Thread "+i); t[i].start(); } for (i=0; i<numThreads; i++) { t[i].join(); } System.out.println("From main: Sum=" + mt.GetSum()); } // end main } // end class
22
22 Data race If you run this program multiple times, you will get different outputs in different runs –A typical symptom of data race –The output depends on how the threads get interleaved when they run –A typical output: : Sum=107927 : Sum=128853 From main: Sum=128853
23
23 More data race example Let us try to write the program for summing an array class MyThread implements Runnable { private int array[]; private long sum; private int numThreads; private int tid; public MyThread (long size, int numThreads) { int i; array = new int[size]; for (i=0; i<size; i++) { array[i] = i; } sum = 0; tid = 0; this.numThreads = numThreads; } // next slide
24
24 More data race example public void run () { int i; int myid = tid++; // data race on tid for (i=myid*(array.length/numThreads); i<(myid+1)*(array.length/numThreads); i++) { sum += array[i]; // data race on sum } public long GetSum() { return sum; } } // end class
25
25 More data race example class MyArraySumDemo { public static void main (String arg[]) throws java.lang.InterruptedException { Integer n = new Integer (arg[0].trim()); long size = n.intValue(); n = new Integer (arg[1].trim()); int numThreads = n.intValue(); int i; MyThread mt = new MyThread(size, numThreads); Thread t[] = new Thread[numThreads]; // next slide
26
26 More data race example for (i=0; i<numThreads; i++) { t[i] = new Thread (mt, "Thread "+i); t[i].start(); } for (i=0; i<numThreads; i++) { t[i].join(); } System.out.println("From main: Sum=" + mt.GetSum() + “[Expected=” + (size*(size-1))/2 + “]”); } // end main } // end class
27
27 More data race example One typical output for size=100000 and numThreads=2 From main: Sum=3692782245[Expected=4999950000] Replacing sum by local sum and accumulating at the end reduces the chance of data race –Reduces the number of executed critical sections: more parallelism –But the program is still buggy and may produce wrong results once in a while
28
28 More data race example class MyThread implements Runnable { private int array[]; private long sum; private int numThreads; private int tid; public MyThread (long size, int numThreads) { int i; array = new int[size]; for (i=0; i<size; i++) { array[i] = i; } sum = 0; this.numThreads = numThreads; } // next slide
29
29 More data race example public void run () { int i; long localSum = 0; // private variable int myid = tid++; // data race on tid for (i=myid*(array.length/numThreads); i<(myid+1)*(array.length/numThreads); i++) { localSum += array[i]; // no data race } sum += localSum; // data race on sum } public long GetSum() { return sum; } } // end class
30
30 More data race example class MyArraySumDemo { public static void main (String arg[]) throws java.lang.InterruptedException { Integer n = new Integer (arg[0].trim()); long size = n.intValue(); n = new Integer (arg[1].trim()); int numThreads = n.intValue(); int i; MyThread mt = new MyThread(size, numThreads); Thread t[] = new Thread[numThreads]; // next slide
31
31 More data race example for (i=0; i<numThreads; i++) { t[i] = new Thread (mt, "Thread "+i); t[i].start(); } for (i=0; i<numThreads; i++) { t[i].join(); } System.out.println("From main: Sum=" + mt.GetSum() + “[Expected=” + (size*(size-1))/2 + “]”); } // end main } // end class
32
32 Fixing data races Every Java object has an in-built lock –We will make use of this to protect the critical sections –For every shared variable that is involved in a data race, we need to create a lock –It is possible to have a single lock to resolve multiple races, but that will limit concurrency (why?) –General solution: create a dummy object and use its lock to resolve a race –Java allows the programmer to mark critical sections with the keyword synchronized
33
33 Fixing array sum class MyThread implements Runnable { private int array[]; private long sum; private int numThreads; private int tid; private Object tidLock; // dummy object private Object sumLock; // dummy object public MyThread (long size, int numThreads, Object tidLock, Object sumLock) { int i; array = new int[size]; for (i=0; i<size; i++) { array[i] = i; } // next slide
34
34 Fixing array sum sum = 0; this.numThreads = numThreads; this.tidLock = tidLock; this.sumLock = sumLock; } public void run () { int i; int myid; synchronized (tidLock) { myid = tid++; } // next slide
35
35 Fixing array sum for (i=myid*(array.length/numThreads); i<(myid+1)*(array.length/numThreads); i++) { synchronized (sumLock) { sum += array[i]; } public long GetSum() { return sum; } } // end class
36
36 Fixing array sum class MyArraySumDemo { public static void main (String arg[]) throws java.lang.InterruptedException { Integer n = new Integer (arg[0].trim()); long size = n.intValue(); n = new Integer (arg[1].trim()); int numThreads = n.intValue(); int i; Object tidLock = new Object(); Object sumLock = new Object(); MyThread mt = new MyThread(size, numThreads, tidLock, sumLock); Thread t[] = new Thread[numThreads]; // next slide
37
37 Fixing array sum for (i=0; i<numThreads; i++) { t[i] = new Thread (mt, "Thread "+i); t[i].start(); } for (i=0; i<numThreads; i++) { t[i].join(); } System.out.println("From main: Sum=" + mt.GetSum() + "[Expected=" + (size*(size-1))/2 + "]"); } // end main } // end class
38
38 Announcements No class tomorrow All the best for the final exam. Bye!
Similar presentations
© 2024 SlidePlayer.com. Inc.
All rights reserved.