CODEWITHSUNDEEP
javaperformancedelayperformance-testingmilliseconds
João Vitor Verona Biazibetti
João Vitor Verona Biazibetti

Reputation: 2243

Big difference on System.currentTimeMillis()

I'm using this little piece of code to test what method is faster:

public void test() {
  long start = System.currentTimeMillis();
  MyDate date = new MyDate();
  int max = 5000;

  for (int i = 0; i < max; i++) {
    Calendar.getInstance().getTimeInMillis(); // <--
  }

  System.out.println("Calendar instance delay: " + (System.currentTimeMillis() - start));
  start = System.currentTimeMillis();

  for (int i = 0; i < max; i++) {
    date.getMillis(); // <--
  }

  System.out.println("My date delay: " + (System.currentTimeMillis() - start));
}

So, basically, I'm trying to compare the performance between Calendar.getInstance().getTimeInMillis() and MyDate.getMillis() (MyDate is a class created by me).

Well, when I ran the code above, the output were:

Calendar instance delay: 413
My date delay: 2

BUT, when I inverted the order (first called MyDate, and after Calendar), I got:

My date delay: 247
Calendar instance delay: 119

I tried using System.nanoTime(), but the same thing occured: the first code to be tested is the one that took longer.

Anyone knows why this difference happens? Or, is there a way to profile codes accurately without using an external profiler application (just pure Java code)?

Thanks.

Upvotes: 1

Views: 2084

Answers (3)

mbelow
mbelow

Reputation: 1133

When measuring performance, keep in mind that the the JVM may eliminate code that has no side effects as part of optimization.

To avoid this, you can do the following:

for (int i = 0; i < max; i++) {
    long a = date.getMillis();
    if (a % 1000 == i) {
        System.out.print(a);
    }
}

Another thing to be aware of is the hotspot-optimization. You should "warm up" your code before measuring.

Upvotes: 1

Avinash Singh
Avinash Singh

Reputation: 3797

There is StopWatch class in commons lang which uses System.nanoTime,

http://commons.apache.org/lang/

org.apache.commons.lang.time.StopWatch

Upvotes: 1

David Schwartz
David Schwartz

Reputation: 182883

To test the performance of "small" code, you have to call it many times. Otherwise, the effects of JIT, caching, and branch prediction mess up your testing. If I call you on the phone to see how long it takes before you answer, I'll get a very different answer if I just called you a few seconds ago.

Upvotes: 4

Related Questions