Re: [nm-wg] standard timestamp format

[Dan Gunter <dkgunter@lbl.gov>]

> > I would suggest that the units be an optional parameter, since they 
> > aren't always needed, and maybe aren't always known, i.e. in the case 
> > of some kind of hardware clock.
>
> If you don't know the clock frequency then the data is meaningless,
> the units should be nanoseconds and you should multiply whatever the
> source clock rate is by some implementation and source specific factor 
> to get to nanoseconds. For the simplest portable java code that would
> be 1000000 to go from milliseconds. Linux doesn't require a hires 
> counter but I think some implementations may have one, Solaris always
> has gethrtime regardless of whether its on SPARC or x86, I know HP-UX, 
> AIX and windows have hires counters.
Not meaningless; if I'm comparing intervals A and B, I don't need any 
units to say that A/B = 1.5. But I see your point that in general we 
want to know what the time intervals really represent. So, sure, let's 
make nanoseconds the standard units. One less piece of metadata to worry 
about. In cases where the units aren't needed, it might as well be 
nanoseconds [ignoring issues with misinterpretation, since this is such 
an edge case anyways].

-Dan

> Adrian