[LEAPSECS] A new use for Pre-1972 UTC

[Rob Seaman]

Poul-Henning Kamp wrote:


> They are trying to make a unique pseudo-random number, not a  

> timestamp.


Well, it's both.  So it turns out that I'm responsible for capturing  
data from a couple of dozen cameras on eleven telescopes on three  
mountaintops in two hemispheres.  Without belaboring all the astronomy  
stuff that has entertained us, lo these many years, there are many  
identifiers and timestamps (and checksums and ...) attached to each  
image file.  One thing we decided to add many years ago was a unique  
ID from each camera.  The format is some variation of:

	<telescope>.<camera>.<iso_8601>.<serial_number>

The precise formatting of the ISO timestamp varies depending on who  
implemented the particular software for that camera, but all are  
standard variations.  The serial number is usually omitted, but it  
permits distinguishing images from cameras that can take very quick  
(or simultaneous) exposures.

I'm guessing that they want to do something similar here.


> To decrease the probability of two persons or programs choosing the

> same number, then include a timestamp in it.

>

> time_t values are readily available in all POSIX filesystems,

> so they use that.


ISO is readily available, too:

	$ date -u +'%Y%m%dT%H%M%S'
	20090217T211132


> This is a variant of the UUID madness that somebody came up with

> because they didn't want to run a registry or use the existing

> well-structured process (ISO OID's) and though that the eventual

> collisions "probably doesn't matter much".


Yeah - it all comes down to use cases and whether the requirements  
address them properly.

Creating an ID that is guaranteed unique is not a trivial task,  
especially if (as one suspects is true here) a central server is out  
of the question.  For example, I control the context for those eleven  
telescopes.  If eleven different organizations were assigning IDs,  
then you have to somehow guarantee that they don't choose the same  
telescope or camera names - because it is a frequent occurrence that  
the timestamps will coincide (no matter what format is chosen).

Hmmm.  I'll have to run the statistics on that - it's the Birthday  
Problem.  A classroom of more than 23 students has a better than even  
chance of some pair of students sharing a birthday.  Similarly, eleven  
telescopes producing a couple of hundred images each per night have a  
high probability of coincidence.

Rob