what is the most precise time I can get in JDK?

[Jeff Hain]

Hi.

I've had the same need (or desire), and made a ThinTime class for it (attached).

Returned time :
- is not much more inaccurate than System.currentTimeMillis(),
- is always ahead of System.currentTimeMillis(),

- is monotonic (unless System.currentTimeMillis() jumps backward),

- should change way more often than System.currentTimeMillis().

NB: It makes use of "raw" AtomicXXX classes, without cache line padding,
so there is room for optimization here, but maybe you are like me and
prefer to wait on @Contended annotation arrival than plague your code
with unreliable padding magic :)


-Jeff


________________________________
 De : Weijun Wang <weijun.wang at oracle.com>
À : David Holmes <David.Holmes at oracle.com> 
Cc : core-libs-dev <core-libs-dev at openjdk.java.net> 
Envoyé le : Mercredi 16 Novembre 2011 9h09
Objet : Re: what is the most precise time I can get in JDK?
 
Hi Davis

Yes, I know nanoTime() is precise, but how can I get a clock time from 
it? I have tried to record a currentTimeMillis() value when the program 
starts and then use the elapse of the nanaTime() to get a current time. 
This will break if the user adjusts the system clock during the program 
execution.

One solution is to keep tracking the changing of both 
currentTimeMillis() and nanoTime(). If the change of one has a 
difference (say, > 1 sec) from the other one, it means a system clock 
change and I can quickly reset my time to currentTimeMillis().

Thanks
Max



On 11/16/2011 03:23 PM, David Holmes wrote:
> Hi Max,
>
> On 16/11/2011 2:55 PM, Weijun Wang wrote:
>> I need a precise time, and is currently using java.util.Date, which knows
>> about milliseconds, but unfortunately the precision is only 10-15
>> milliseconds on a Windows.
>>
>> In fact, I don't really need it to be so correct. My requirements are:
>>
>> 1. It's enough correct, say, at least as correct as Date.
>> 2. It's precise in a relative sense, i.e. it changes fast
>> 3. It should be monotonic, i.e. it grows, unless the user adjusts the
>> system
>> clock
>
> There are only two time source available:
>
> 1. The time-of-day clock
>
> This is what Date reports and is also what System.currentTimeMillis
> reports. It only has millisecond precision. It's rate of update is
> dependent on the OS - for Windows that is typically every 10ms or every
> 15ms depending on version.
>
> 2. The high resolution time source
>
> This is what System.nanoTime reports. It has nanosecond precision, but
> again depending on the OS it's resolution (update rate) will vary. The
> update rate should easily be in the tens of microseconds. It should be
> monotonic non-decreasing but it is not connected to the time-of-day
> clock (and so should not be affected by any changes therein).
>
> I have an old blog entry on this:
>
> http://blogs.oracle.com/dholmes/entry/inside_the_hotspot_vm_clocks
>
> David
> -----