I must create simple application with threads. Task: faster calculation than serial processing. I must use methods notify() or notifyAll(), wait(), interrupt() and operator synchronized.
I tried solve this by example from book. This example is typical producent consumer task. But parallel threads calculating is slower then serial.
Class with notify(), wait() and synchronized:
public class Reader {
private boolean isRead = false;
private boolean isFileEnd = false;
private int value;
private int[] pole;
private int pocitadlo=0;
public Reader(int[]pole) {
this.pole=pole;
}
synchronized public void loadValue() {
while (isRead == true) {
try {
wait();
}
catch (InterruptedException e) { }
}
if (pocitadlo<pole.length) {
value = pole[pocitadlo];
pocitadlo++;
}
else {
isFileEnd = true;
Thread.currentThread().interrupt();
}
isRead = true;
notifyAll();
}
synchronized public int getValue() {
while (isRead == false) {
try {
wait();
}
catch (InterruptedException e) { }
}
isRead = false;
if (isFileEnd == false) {
notifyAll();
return value;
}
else {
Thread.currentThread().interrupt();
return 0;
}
}
}
Producer class:
public class Producent extends Thread {
private Reader reader;
public Producent(Reader reader) {
this.reader = reader;
}
@Override
public void run() {
while (interrupted() == false) {
reader.loadValue();
}
}
}
Consumer class:
public class Konzument extends Thread {
private Reader reader;
private double sum = 0;
public Konzument(Reader reader) {
this.reader = reader;
}
@Override
public void run() {
int number;
while (true) {
number = reader.getValue();
if (interrupted() == false)
sum += Math.sqrt(number);
else
break;
}
}
public double getSum(){
return sum;
}
}
And main class code (I know that I can create objects in cycles for bigger comfort but this is only example):
public class PARPV2 {
/**
* @param args the command line arguments
*/
public static void main(String[] args) {
Watch watch=new Watch();
double suma=0;
int size=1000000;
int[]array1=new int[size];
int[]array2=new int[size];
int[]array3=new int[size];
int[]array4=new int[size];
int[]array5=new int[size];
generate(array1);
generate(array2);
generate(array3);
generate(array4);
generate(array5);
Reader reader1=new Reader(array1);
Reader reader2=new Reader(array2);
Reader reader3=new Reader(array3);
Reader reader4=new Reader(array4);
Reader reader5=new Reader(array5);
Producent p1=new Producent(reader1);
Konzument k1=new Konzument(reader1);
Producent p2=new Producent(reader2);
Konzument k2=new Konzument(reader2);
Producent p3=new Producent(reader3);
Konzument k3=new Konzument(reader3);
Producent p4=new Producent(reader4);
Konzument k4=new Konzument(reader4);
Producent p5=new Producent(reader5);
Konzument k5=new Konzument(reader5);
watch.start();
p1.start();
k1.start();
p2.start();
k2.start();
p3.start();
k3.start();
p4.start();
k4.start();
p5.start();
k5.start();
try {
p1.join();
k1.join();
p2.join();
k2.join();
p3.join();
k3.join();
p4.join();
k4.join();
p5.join();
k5.join();
suma=k1.getSum()+k2.getSum()+k3.getSum()+k4.getSum()+k5.getSum();
System.out.println("paralelne: "+watch.stop(false));
System.out.println("suma: "+suma);
} catch (InterruptedException ex) {
Logger.getLogger(PARPV2.class.getName()).log(Level.SEVERE, null, ex);
}
double sum1=0;
double sum2=0;
double sum3=0;
double sum4=0;
double sum5=0;
watch.start();
for (int i = 0; i < array1.length; i++) {
sum1+=Math.sqrt(array1[i]);
}
for (int i = 0; i < array2.length; i++) {
sum2+=Math.sqrt(array2[i]);
}
for (int i = 0; i < array3.length; i++) {
sum3+=Math.sqrt(array3[i]);
}
for (int i = 0; i < array4.length; i++) {
sum4+=Math.sqrt(array4[i]);
}
for (int i = 0; i < array5.length; i++) {
sum5+=Math.sqrt(array5[i]);
}
suma=sum1+sum2+sum3+sum4+sum5;
System.out.println("serial: "+watch.stop(false));
System.out.println("suma: "+suma);
}
public static void generate(int[]array){
Random r=new Random();
for (int i = 0; i < array.length; i++) {
array[i]=r.nextInt(100);
}
}
}
When I run this program, serial computing is much faster than parallel. But I have dual core and I thing that parallel calculation must be faster. I am not experienced in threads and maybe my example is dummy...Please, where can be a problem? Thanks
If you want faster execution using Threads, I think that you should use a single class that will accomplish the same thing as serial, something like that :
public class Calc extends Thread {
private int[] array;
private double sum;
public Calc(int[] array) {
this.array = array;
this.sum = 0;
}
@Override
public void run() {
for (int i = 0; i < array.length; i++) {
sum += Math.sqrt(array[i]);
}
}
public double getSum() {
return this.sum;
}
}
In this way, parallel computing is 2 times faster than serial coputing. I think the pattern Producer-Consummer you applied can't be faster than serial because you make way more instructions and breaks in the code by :
Maybe the producer should be the one who generates random number, and the consummer the one using these values ?
(And please use while(isRead) and ! operator for the contrary instead of those ugly while(isRead == false) : this is more readable ;))