System.nanoTime(): good or evil?

Newbie & Debugging Questions
#21 
public class MRes{
	public static void main(String[]args){
		long smallestDelta=Long.MAX_VALUE;
		long start=System.currentTimeMillis();
		long current=start;
		long last=start;
		long delta;
		System.out.println("testing...");
		while(current-start<5000){
			current=System.currentTimeMillis();
			delta=current-last;
			if(delta!=0&&delta<smallestDelta)
				smallestDelta=delta;
			last=current;
		}
		System.out.println(System.getProperty("os.name")+" ["+
			System.getProperty("os.version")+"]["+System.getProperty("os.arch")+"]"
		);
		System.out.println("smallest delta: "+smallestDelta+"msec");
	}
}

gives me:

testing...
Windows 2000 [5.0][x86]
smallest delta: 10msec
#22

testing…
Windows XP [5.1][x86]
smallest delta: 15msec

testing…
Mac OS X [10.4.9][i386]
smallest delta: 1msec

#23

The Mac result was as expected. The XP one really baffles me tho.

#24

Same here:

testing…
Windows XP [5.1][x86]
smallest delta: 15msec

#25

and here as well:

testing…
Windows XP [5.1][x86]
smallest delta: 15msec

#26 
testing...
Linux [2.6.20-15-generic][i386]
smallest delta: 1msec

Hippies are good at some things I suppose…

#27

Does anybody know why currentTimeMillis on Windows is so rough?

Maybe we should submit an enhancement request to have 1ms precision on Windows too (it’s not a bug since currentTimeMillis doesn’t guarantee anything).

#28

Does anybody know why currentTimeMillis on Windows is so rough?

It uses the OS’ tick counter, whose resolution depends on the OS.

Here is some overview:
http://www.gamedev.net/reference/programming/features/timing/

As you can see TGT has (had?) a few issues, too. However, QPC has a lot more.

#29

Do you have any useful test to verify nanoTime() ?
When I used nanoTime, it always worked out well, but I never ran more than a few seconds. I dont know if it can
make trouble when running for minutes / hours, if it has temporary errors or so.

-JAW

#30

did you read the whole thread? you realize that it’s only a problem on multi-core cpus? Sorry if you are talking about them, I got impression that you aren’t.

#31

you realize that it’s only a problem on multi-core cpus?

Heh. That would be nice. But there are a dozen different issues with QPC.

I got a single core and QPC is unusable (QPC leaping).

#32

When using System.nanoTime() on my computer with a single-core CPU, the game I wrote would suddenly pause for about half a second every once in a while (about once per level). When using the millisecond timer instead, this problem stopped.

So System.nanoTime() has problems on single-core processors (or at least on mine) as well. System.nanoTime() is evil.

#33

I’ve talked with some other (non-java) developers about the issue, and all of them (okay, they were just 3 people :)) use 1ms timing instead of QPC for exactly the same reasons.

#34

damn Sun developers, why did they used damn QPC :slight_smile: … is there a RFE or bug report on this? It has to be.
I’ve heard LWJGL times is good, why don’t they just use that, or whatever it’s based on.

#35

…I suppose because the RFE it was based on required high precision, and LWJGL’s timer is just 1ms.
I’d rather see a currentTimeMillis implementation with 1ms precision on Windows.

#36

Somehow, I would like to see that. However, since TGT is something completely different (and may not work properly on some systems, too) a new method would be required if you don’t want to create regressions.

The most logical name would be…
System.milliTime() (although… I would like msecTime() more ;))