Performance test:singleton class with and without double check locking - performance

Question

I have two implementations of singleton classes
public class Test2 {
private static Test2 _instance=new Test2();
private Test2(){
}
public static synchronized Test2 getInstance(){
if(_instance == null){
_instance = new Test2();
}
return _instance;
}
}
And:
public class TestSingleton {
private static TestSingleton _instance=new TestSingleton();
private TestSingleton(){
}
public static TestSingleton getInstance(){
if (_instance == null) {
synchronized (TestSingleton.class) {
if (_instance == null) {
_instance = new TestSingleton();
}
}
}
return _instance;
}
I want to parametrize my finding in terms of time taken, what I did is this:
Callable<Long> task = new Callable<Long>() {
#Override
public Long call() throws Exception {
long start = System.nanoTime();
**TestSingleton.getInstance();**
long end = System.nanoTime();
return end - start;
}
};
for (int i = 0; i < 100000; i++) {
futList.add(es1.submit(task));
}
for (Future<Long> fut : futList) {
try {
totalTime1.getAndAdd(fut.get());
} catch (InterruptedException e) {
// TODO Auto-generated catch block
e.printStackTrace();
} catch (ExecutionException e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
}
System.out.println("Time taken by S1 " + totalTime1.get());
.
.
ExecutorService es2 = Executors.newFixedThreadPool(threadpool);
Callable<Long> task1 = new Callable<Long>() {
#Override
public Long call() throws Exception {
long start = System.nanoTime();
Test2.getInstance();
long end = System.nanoTime();
return end - start;
}
};
for (int i = 0; i < 100000; i++) {
futList1.add(es2.submit(task1));
}
for (Future<Long> fut : futList1) {
try {
totalTime2.getAndAdd(fut.get());
} catch (InterruptedException e) {
// TODO Auto-generated catch block
e.printStackTrace();
} catch (ExecutionException e) {
// TODO Auto-generated catch block
e.printStackTrace();
}
}
System.out.println("Time taken by S2 " + totalTime2.get());
The results I got is:
Time taken by S1 4636498
Time taken by S2 5127865
First question is this the correct approach? and second even if I comment the getinstances method in both the call(), I get different times of execution of two identical blocks:
Time taken by S1 1506640
Time taken by S2 2156172

Answer 1

Don't measure each execution and sum the times, there will be too much inaccuracy in the individual measurements. Instead, get start time, execute 100000 times, get end time. Also, execute a few 1000 times before you start measuring to avoid skewing by start-up costs.