perfSONAR User Q&A and Other Discussion

[Tim Chown <>]

> On 15 Nov 2022, at 15:43, Raul Lopes <> wrote:
> 
> Chrony should help withh PTP, according to Red Hat. However, I think that 
> perfSOAR still wants ntpd, right?
> 
> I would hope that PTP might drive away a lot of "negative latency" we see 
> in some pS dashboards.

It would be interesting to dive into the data where there are examples of 
this, presumably for very low OWD paths, sub 2-3ms, or even lower than that.  
Is it NTP not being accurate enough, or some issue with NTP config or the 
network paths involved.

Some data driven analysis of perhaps some of the WLCG the perfSONR latency 
data might be enlightening.

Tim

> 
> Raul
> 
>  From: Shawn McKee <>
> Sent: 15 November 2022 15:22
> To: Tim Chown <>
> Cc: Szymon Trocha <>; Raul Lopes 
> <>;  
> <>
> Subject: Re: [perfsonar-user] PTP supported?
>  Hi Everyone,
> 
> I would second (or third) the enhancement to support PTP.  We just got our 
> PTP antenna installed for AGLT2 a few weeks ago.  We have PTP capable Dell 
> switches using PTP to hold times.
> My challenge is to get our perfSONAR instances to use it.
> 
> Shawn
> 
> On Tue, Nov 15, 2022 at 6:38 AM Tim Chown <> 
> wrote:
> > On 14 Nov 2022, at 15:03, Szymon Trocha <> 
> > wrote:
> > 
> > Hi Raul,
> > 
> > W dniu 10.11.2022 o 19:35, Raul Lopes (via perfsonar-user Mailing List) 
> > pisze:
> >> Dear perfsonar team,
> >> 
> >> Is PTP supported in perfSONAR? 
> >> 
> > I addition to what Ed replied - I'm wondering about the rationale behind 
> > your question. Is PTP configured and used in JISC? Is it going to be your 
> > replacement of NTP?
> > 
> > (https://docs.perfsonar.net/FAQ.html#q-does-perfsonar-support-ptp)
> 
> No, at this stage there’s no PTP service on Janet.  But other NRENs are 
> starting to offer one, and we’re in the process of looking at use cases.  
> One may well be more accurate OWD measurements.
> 
> One thing the perfSONAr community could do, or a researcher in it, is look 
> at the existing latency measurements each way between measurement points 
> and see how they correlate, and whether there is clock drift.  My feeling 
> is that mostly in R&E networks there’s ~1ms accuracy based on existing NTP 
> driven timing.
> 
> The current perfSONAr docs say PTP is not supported due to the cost of 
> deploying it.  I think that has changed.
> 
> Tim--
> To unsubscribe from this list: 
> https://lists.internet2.edu/sympa/signoff/perfsonar-user