[ntp:questions] Re: isolated network + accuracy + next_day

Richard B. Gilbert rgilbert88 at comcast.net
Wed Dec 15 14:27:22 UTC 2004


Alaios wrote:

>Thx a lot... I will ask my final question.. What isthe
>accuracy that can be achieved in an isolated network?
>I must say something to my boss. Why the accuracy is
>so small when we have a Fast Ethernet with a
>full-duplex switch for all the clients?
>
>Also the accuracy what is the accuracy that the
>internet provides? What should i answer him... We have
>no GPS os other hardware can we achieve accuracy below
>the 1ms....?? I suppose no
>
>
>--- "Richard B. Gilbert" <rgilbert88 at comcast.net>
>wrote:
>
>  
>
>>Bjorn Gabrielsson wrote:
>>
>>    
>>
>>>Bjorn Gabrielsson <bg at lysator.liu.se> writes:
>>>
>>> 
>>>
>>>      
>>>
>>>>"Richard B. Gilbert" <rgilbert88 at comcast.net>
>>>>        
>>>>
>>writes:
>>    
>>
>>>>   
>>>>
>>>>        
>>>>
>>>>>Alaios wrote:
>>>>>     
>>>>>
>>>>>          
>>>>>
>>>>>>ntpq -c pe ripebox.grnet.gr
>>>>>>   remote           refid      st t when poll
>>>>>>            
>>>>>>
>>reach delay   offset
>>    
>>
>>>>>>jitter
>>>>>>            
>>>>>>
>>>>>==============================================================================
>>>>>          
>>>>>
>>>>>>*GPS_RIPENCC(0)  .GPS.            0 l    3   64 
>>>>>>            
>>>>>>
>>377  0.000   -0.002
>>    
>>
>>>>>>0.004
>>>>>>
>>>>>>       
>>>>>>
>>>>>>            
>>>>>>
>>>>>That's two microseconds!  It's "valid" only as of
>>>>>          
>>>>>
>>the instant it was
>>    
>>
>>>>>estimated.   The estimate will be updated each
>>>>>          
>>>>>
>>time the server is
>>    
>>
>>>>>polled.
>>>>>     
>>>>>
>>>>>          
>>>>>
>>>>No he was asking 'ripebox.grnet.gr' good its
>>>>        
>>>>
>>_local_ time is wrt the time it
>>    
>>
>>>>receives from its Trimble GPS receiver.
>>>>   
>>>>
>>>>        
>>>>
>>>Hmmm should not be posting when in a hurry...
>>>      
>>>
>>Insert a 'how' at suitable
>>    
>>
>>>place to make previous post parseable. *Sorry*
>>> 
>>>
>>>      
>>>
>>And the local clock is two microseconds off. 
>>_______________________________________________
>>questions mailing list
>>questions at lists.ntp.isc.org
>>https://lists.ntp.isc.org/mailman/listinfo/questions
>>
>>    
>>
>
>
>__________________________________________________
>Do You Yahoo!?
>Tired of spam?  Yahoo! Mail has the best spam protection around 
>http://mail.yahoo.com 
>  
>
Definitions first.

Accuracy:  The magnitude of the difference between your clock and the 
the "correct" UTC time.   You can't know this without knowing what UTC 
is and, in an isolated network with no hardware reference clock, you 
have no way of knowing what the correct time is.  With good luck and 
good reflexes you might be able to set the clock on your server to 
within plus/minus one second using your wristwatch or your cellular 
phone (the best source)

Precision:  The granularity with which you can set or read the clock.  
How much time does the least significant digit (bit) represent?  
Determined by the hardware and the O/S.

Closeness or tightness of synchronization:  the standard deviation  of  
the  clients with respect to the server.

The closeness of synchronization is the only thing that you have much 
control of.  That will depend primarily on the stability of your 
server's clock.  The typical unsynchronized computer clock does not have 
good stability.  It depends primarily on the temperature of the clock 
circuitry which, in turn, depends on the room temperature.  There are 
also dependencies on the supply voltage, the age of the crystal, etc.

I have seen ten unsynchronized PC clocks develop a spread of twelve 
hours over the course of a year!!!   These PCs were all the same make 
and model and were originally set to the correct time (from my wrist 
watch) when I installed them.  The PCs were running Solaris 8 Intel 
Platform Edition.   When I configured NTP using a server on the local 
network (a DEC Alpha server 4000 running VMS V7.2-1), which in turn was 
synchronized via the internet, they synched up to within 100 
microseconds of each other and the server.  The server was usually 
within 10 milliseconds of the correct time!

If your server is not synchronized you are, in effect, standing on a 
randomly moving platform and shooting at a randomly moving target.  I 
don't know if anyone has ever tried to make a scientific study of the 
behavior of such a system.  I'd say that an accuracy of 1ms could be 
achieved only by random chance!  Synchronization within 1 ms might be 
achievable; you could try it and see.




More information about the questions mailing list