Who can explain this?

[Stephane Clinckart]

Who can expalin this ???

final ThreadMXBean threadMXBean = ManagementFactory.getThreadMXBean();
final long startNanoSystem = System.nanoTime();
final long start = threadMXBean.getCurrentThreadCpuTime();
long stop;
while ((stop = threadMXBean.getCurrentThreadCpuTime()) == start);
final long stopNanoSystem = System.nanoTime();
System.out.println("Resolution            : " + (stop - start) + "ns - " + ((stop - start)/(1000L*1000L)) + "ms");
System.out.println("Resolution Nano System: " + (stopNanoSystem - startNanoSystem) + "ns - " + ((stopNanoSystem - startNanoSystem)/(1000L*1000L)) + "ms");

Resolution : 15600100ns - 15ms
Resolution Nano System: 810029ns - 0ms

"Resolution" should be lower than "Resolution Nano System" but it is not the case !

Thanks for help.

PS: The original code is from this blog: http://snippets.hoydaa.org/simple-stopwatch-to-calculate-actual-thread-time

[Paul Clapham]

Looking at the values of those four numbers, instead of just looking at the difference between pairs of them, might illuminate the question. Provided you can replicate the phenomenon reliably, that is.

[William Brogden]

Just read the JavaDocs for java.lang.System and you will see that nanoTime has lots of limitations.

Bill

[Darryl Burke]

Also read UseAMeaningfulSubjectLine

[Campbell Ritchie]

Last time I used nanoTime() a lot, it appeared to show the execution time to the nearest ~0.4µs. That will obviously differ from machine to machine.

[Stephane Clinckart]

I'm not sure that I explain well.

What I don't understand is that Sustem.getNano is arround the other way to get the duration.

It looks like both methods are not using the same time referencial internaly.

start1 = getTime()
start2 = getCPUTime()

stop2 = getCPUTime()
stop1 = getTime()

Normally (stop1-start1) should be higger than (stop2-start2)

But it is not the case.

Even if I try in a loop.

Any idea what cause this ???

[Jesper de Jong]

Those timers are not very precise; to start with, their resolution depends on the operating system you're running on and can be as high as 50 ms or so. I also remember reading somewhere that under some circumstances the clock could even seem to be running backwards in time from one moment to the next.

Don't count on these timers to measure very short amounts of time very precisely.