perfSONAR User Q&A and Other Discussion

[Jason Zurawski <>]

FWIW, the horrible perfSONAR webmaster again forgot to migrate a page from 
the old site that talked about NTP.  He will be punished, severely, for his 
crimes.  Here it is now, perhaps it will be semi useful:

http://www.perfsonar.net/deploy/timekeeping-with-ntp/

Thanks;

-jason

On Dec 5, 2014, at 2:20 PM, Brian Candler 
<>
 wrote:

> On 05/12/2014 18:32, Eli Dart wrote:
>> 
>> Just out of curiosity, why would you trust ping to measure latency if 
>> you're not satisfied with OWAMP?
>> 
> Because the round-trip latency measurement doesn't depend on clock skew.
> 
>> If NTP clock skew is causing you problems, I'm guessing you're measuring 
>> latency over very short distances, but I'd like to ask before I assume.  
>> What sort of network are you trying to measure?  I apologize if I've 
>> missed it in earlier threads.
>> 
> Yes, it's short distances - right now it's in the office for test, and when 
> deployed it will be over a few hops of fibre.
> 
> However I've just discovered that the office ISP appears to be blocking NTP 
> responses - probably a misguided attempt to protect against NTP DoS attacks 
> - so maybe NTP should be able to achieve better than I have been seeing 
> once the boxes go into the field. I'll also try to give the ISP some clue.
> 
> Regards,
> 
> Brian.

On Dec 5, 2014, at 2:21 PM, Alan Whinery 
<>
 wrote:

> Because, as he said, owamp reports between two participants whose path
> latency is similar or less than a typical NTP clock correction is more
> noise than signal.
> 
> I wouldn't suggest hiding the latency, but in close-quarters owamp, the
> loss really is the useful part.
> 
> Brian - As for clock discipline standards, you may want to look at
> EndRun CDMA devices. They usually don't require the antenna-on-roof
> aspect that a GPS install does, but provide similar precision.
> 
> http://endruntechnologies.com/time-frequency-reference-cdma.htm
> 
> http://net.its.hawaii.edu/network-performance/using-praecis/
> 
> Of course, you did say "not-too-expensive". In 2010, I paid $1,375.00
> each for Praecis II.
> 
> On 12/5/2014 8:32 AM, Eli Dart wrote:
>> Hi Brian,
>> 
>> Just out of curiosity, why would you trust ping to measure latency if
>> you're not satisfied with OWAMP?
>> 
>> If NTP clock skew is causing you problems, I'm guessing you're measuring
>> latency over very short distances, but I'd like to ask before I assume.
>> What sort of network are you trying to measure?  I apologize if I've missed
>> it in earlier threads.
>> 
>> Eli
>> 
>> 
>> On Fri, Dec 5, 2014 at 10:18 AM, Brian Candler 
>> <>
>>  wrote:
>> 
>>> I'm finding that the latencies I'm trying to measure are less than the
>>> differences in NTP-synced time, and this leads to very misleading graphs.
>>> 
>>> 1. Does anyone have any suggestions for a good, but not-too-expensive, GPS
>>> time source to use with perfsonar/CentOS?
>>> 
>>> 2. Is it possible to configure perfsonar to graph the sensitive packet
>>> loss from owamp, but hide the latency? (I can always use ping for
>>> round-trip latency)
>>> 
>>> Thanks,
>>> 
>>> Brian.