[ntp:questions] NNTP server causing large jumps in time

David L. Mills mills at udel.edu
Fri May 2 18:21:19 UTC 2008


Greg,

Yes, I know that case. There are two reasons why iburst is not the 
default with reference clocks. First, some drivers make use of the 
iburst code to do exactly what iburst does but with special application 
to the clock protocol itself. For instance, some drivers (WWVB, etc.) 
use a burst every minute to fill the median filter. In other words, a 
statistically clean estimate is not ready until the filter has been 
filled and the median/trimmed average has been computed.

Second, some drivers (audio WWV, etc.) cannot deliver any sample at all 
until after some minutes gringing through DSP algorithms and maximum 
likelihood claptrap. GPS receivers are a different matter and presumably 
could opeate with iburst, although as in the WWVB case a really good 
timecode estiamte is pssible only after the median filter. With the 
Spectracom GPS driver and PPS atom driver, the PPS starts running when 
the first timecode sample is received. This generally results in the 
GPS/PPS combination coming up before external sources. For drivers that 
don't use the atom driver, each case might differ.

Dave

Greg Dowd wrote:

> A good example of unexpected behavior is that Dave has implemented the
> notorious KoD server quench code.  As he has tweaked it, I know there
> are some cases where iburst could result in KoD messages from server.  
> 
> As for ref clocks, I think iburst is a great option and I wish it were
> supported.  While oscillators may still be warming up, or gps might not
> be locked, the clock driver can choose to return an error but will quite
> often still be FAR more accurate than a network source.  I know in some
> of my boxes that have gps for primary and multiple redundant network
> servers for backup can often select the network source first (especially
> if the delay causes clustering that pushes them away from GPS!) and then
> switch over to gps later.
> 
> 
> 
>   
> Greg Dowd
> gdowd at symmetricom dot com (antispam format)
> Symmetricom, Inc.
> www.symmetricom.com
> "Everything should be made as simple as possible, but no simpler" Albert
> Einstein
>  
> 
> 
>>-----Original Message-----
>>From: questions-bounces+gdowd=symmetricom.com at lists.ntp.org 
>>[mailto:questions-bounces+gdowd=symmetricom.com at lists.ntp.org]
>> On Behalf Of Kevin Oberman
>>Sent: Friday, May 02, 2008 8:10 AM
>>To: Ryan Malayter
>>Cc: questions at lists.ntp.org
>>Subject: Re: [ntp:questions] NNTP server causing large jumps in time
>>
>>
>>>Date: Fri, 2 May 2008 07:37:32 -0500
>>>From: "Ryan Malayter" <malayter at gmail.com>
>>>Sender: questions-bounces+oberman=es.net at lists.ntp.org
>>>
>>>
>>>On Thu, May 1, 2008 at 12:25 PM, Steve Kostecke 
>>
>><kostecke at ntp.org> wrote:
>>
>>>>You should always append iburst to your server lines.
>>>
>>>Aussuming that were true, why isn't iburst the default? You 
>>
>>would then 
>>
>>>of course have to add "noiburst" to turn it off...
>>
>>I suspect it is because iburst is relatively new and changing 
>>defaults is something that is usually done with great 
>>deliberation and only after people are REALLY sure that it is 
>>the right answer, especially when it only results in an 
>>improvement in startup sync and does not fix anything that 
>>was failing.
>>
>>I can't think of cases where 'iburst' would break things, 
>>though some special cases might be better off without it, but 
>>for 'normal'
>>configurations using the network to chime with servers. I 
>>would not use it on a reference clock, but I have not given a 
>>lot of thought to what the effect of this might be.
>>--
>>R. Kevin Oberman, Network Engineer
>>Energy Sciences Network (ESnet)
>>Ernest O. Lawrence Berkeley National Laboratory (Berkeley Lab)
>>E-mail: oberman at es.net			Phone: +1 510 486-8634
>>Key fingerprint:059B 2DDF 031C 9BA3 14A4  EADA 927D EBB3 987B 3751
>>




More information about the questions mailing list