[NTLUG:Discuss] Slightly OT: Why would I care about Leap Seconds

[Leroy Tennison]

Fred James wrote:
> Chris Cox wrote:
>
>   
>> Leroy Tennison wrote:
>>  
>>
>>     
>>> Fred James wrote:
>>>    
>>>
>>>       
>>>> All
>>>> Why would I care about Leap Seconds?  In the first paragraph at 
>>>> <http://cspry.co.uk/computing/Indy_admin/TIMEZONE.html> it says: 
>>>> "However, `TIMEZONE', as implemented in this way in IRIX, does not take 
>>>> into account leap seconds, nor changes made by the European Union and 
>>>> other bodies, in daylight savings time. The TIMEZONE settings have to be 
>>>> altered manually every year, to update them, as changeovers are often 
>>>> variable.**"  Do I care, and if so why?  No, I am not in the UK or 
>>>> Europe - I am in Texas.  Thank you in advance for any help you may be 
>>>> able to offer.
>>>> Regards
>>>> Fred James
>>>>
>>>> PS:  I am using TZ=CST6CDT5,M3.2.0/2,M11.1.0/2 to deal with the new 
>>>> rules and so far testing has shown it to work on SGI IRIX 6.*, and on 
>>>> Unisys UNIX SVR4 release 2.  I shall be testing on some Linux boxes soon.
>>>>
>>>>  
>>>>      
>>>>
>>>>         
>>> Unless you are doing something esoteric and needing to keep in sync. 
>>> with other machines that are "leap second aware" I would ask the 
>>> question "Who cares about a second of deviation or even a minute?" Sure, 
>>> with NTP you can stay within a second or two of UCT and that's not a bad 
>>> idea but somewhere the "Get a life" sanity check needs to be introduced. 
>>> Of course, with a question like this, "Ask five experts, get six 
>>> answers" is probably going to apply.
>>>    
>>>
>>>       
>> Actually, using NTP you should be able to sync well within microseconds
>> of time from the server (close enough for almost anything to not care).
>>
>> And time sync is VERY important.  There are many critical infrastructure
>> services that will not operate correctly unless all hosts are in
>> sync with each other.
>>  
>>
>>     
> All
> Thank you all for your replies.
>
> I have four servers working as "time slaves" to another box acting as 
> "time master", so those boxes are all in sync with each other, and I 
> manually correct the master whenever it seems to be floating more than a 
> minute or so off <www.time.gov> - not really much floating going on.  
> The servers (out of my control) we communicate with are a mix of UNIX, 
> Linux, and Windows boxes, and so far not one has complained - when I 
> have checked I have found the Windows boxes to be notoriously off, and 
> the UNIX boxes to be close to, but usually in disagreement with 
> <www.time.gov> by a minute or two themselves, and not in agreement with 
> each other or my boxes.  So I guess you could say close is good enough here.
>
> But I am interested none the less in why I might care (as in another 
> situation).  Perhaps you could expand on the the "There are many 
> critical infrastructure services that will not operate correctly unless 
> all hosts are in sync with each other" a little bit, please.  Thank you 
> in advance for any help you may be able to offer.
> Regards
> Fred James
>
>
> _______________________________________________
> http://www.ntlug.org/mailman/listinfo/discuss
>
>   
What comes to mind immediately is the time system itself.  I have read 
in some of the NTP-related documents that propagation delay is taken 
into account between stratums.  Certainly if that is being considered 
then leap-seconds need to be accounted for. 

A more "real-life" example would be Novell NDS which uses time as the 
"tie breaker" in deciding what update to apply in the event of 
collisions (since updates can come from multiple sources - there is no 
master update source).  It also uses time in it's synchronization 
processing (updates can be applied to multiple databases which in turn 
need to be kept in sync.).  However, in this situation, time 
synchronization within about two seconds is adequate.

It should be noted that, with a number of systems where time 
synchronization is required, it is not critical that the time be 
accurate - only that the agreed-upon time be within a synchronization 
range.  As long as all systems agree closely enough with a common but 
erroneous time then everything is OK - this is the old precision versus 
accuracy issue.  Of course, if the time also happens to be accurate then 
it is easier to add new devices, deal with larger environments and 
incorporate dis-similar systems...

http://www.endruntechnologies.com/time-server-why.htm lists some 
examples of where time synchronization is critical but probably not to 
the leap second level.  Another interesting (and short) reference is 
http://www.raltron.com/products/pdfspecs/sync_an02-StratumLevelDefined.pdf.  
It contains a statement that 3.7 x 10^-7 is not acceptable for SONET.

An area where I would think it was critical is in cooperative efforts to 
do calculations in the realm of astronomy.  If you are working with 
others trying to plot planetary positions in time past and/or future 
then certainly taking every possible source of difference into account 
would be important.  Just went and did a Goggle query for "leap second" 
and what seems to be showing up is NTP (the fact that a number of NTP 
servers aren't honoring this adjustment) and issues with astronomy.  
Worth noting is that leap seconds is something that apparently is voted 
upon...  Also worth noting is that there have been 23 of them since 1972.

Back to "reality", I would be much more concerned about the coming 
daylight savings time change where by law its beginning and end was 
moved by days and you could be out of synchronization with "real" time 
by an hour for an extended period of time if you didn't take this into 
account.  On this subject, the Microsoft and NetWare worlds have patches 
to apply for this issue, what is the situation with Linux?