linuxptp-users

[Linuxptp-users] Interpretation of ptp4l results

[Chandra M. <sma...@al...>]

Hi Friends,

I would like to double-check whether my following interpretation is correct. Please elucidate me.

For the following ptp4l output, the ptp4l reports rms value and the max deviation from the mean offset. Thus, if I have to report the worstcase accuracy, I take the (rms+max). For example, , looking at the first line, I see that the maximum offset uncertainty/inaccuracy  observed is 2 +7 = 9ns. The maximum frequency uncertainty is (156403 + 7) pbb.


ptp4l[833.369]: rms    2 max    7 freq -156403 +/-   7 delay    28 +/-   0

ptp4l[834.399]: rms    2 max    7 freq -156403 +/-   8 delay    29 +/-   0

ptp4l[835.428]: rms    2 max    7 freq -156403 +/-   7

ptp4l[836.457]: rms    2 max    6 freq -156403 +/-   7 delay    28 +/-   0

ptp4l[837.487]: rms    2 max    6 freq -156403 +/-   7 delay    28 +/-   0

ptp4l[838.516]: rms    2 max    6 freq -156403 +/-   7 delay    27 +/-   0

ptp4l[839.549]: rms    2 max    7 freq -156403 +/-   7

ptp4l[840.583]: rms    2 max    8 freq -156403 +/-   8 delay    29 +/-   0


Thanking you in anticipation,
Regards,
Chandra

(c) : +60.175508142
(O): +60.4.636.6412

"Knowledge speaks, Wisdom listens"



________________________________

Confidentiality Notice.
This message may contain information that is confidential or otherwise protected from disclosure. If you are not the intended recipient, you are hereby notified that any use, disclosure, dissemination, distribution, or copying of this message, or any attachments, is strictly prohibited. If you have received this message in error, please advise the sender by reply e-mail, and delete the message and any attachments. Thank you.

Re: [Linuxptp-users] Interpretation of ptp4l results

[Miroslav L. <mli...@re...>]

On Fri, Jun 19, 2015 at 07:53:52AM +0000, Chandra Mallela wrote:
> 
> 
> Hi Friends,
> 
> I would like to double-check whether my following interpretation is correct. Please elucidate me.
> 
> For the following ptp4l output, the ptp4l reports rms value and the max deviation from the mean offset. Thus, if I have to report the worstcase accuracy, I take the (rms+max). For example, , looking at the first line, I see that the maximum offset uncertainty/inaccuracy  observed is 2 +7 = 9ns. The maximum frequency uncertainty is (156403 + 7) pbb.

The max value is just the maximum measured offset in that interval. It
doesn't mean it's also the worst accuracy of the clock. There could be
an asymmetry in the network delay and ptp4l wouldn't know the clock is
actually behind or ahead of the time source. To evaluate the accuracy
you would to measure the error independendly from PTP, for example by
comparing PPS signal generated by the synchronized clock with the
reference on a scope.

-- 
Miroslav Lichvar

Re: [Linuxptp-users] Interpretation of ptp4l results

[Chandra M. <sma...@al...>]

Hi Miroslav,

My question is about mathematical interpretation of the output from the ptp4l stack. If I see the output like the one below, is it fair to assume that the 'rms' (root mean squre) is the mean offset of all the sample offsets observed in the interval and 'max' is the maximum deviation observed from the rms value? I am likening it to kinda gaussian/normal distribution but without sign interpretation. Otherwise, I am fully in line with what you have mentioned below.

ptp4l[833.369]: rms    2 max    7 freq -156403 +/-   7 delay    28 +/-   0

Thanking you in anticipation,
Regards,
Chandra

(c) : +60.175508142
(O): +60.4.636.6412

"Knowledge speaks, Wisdom listens"



-----Original Message-----
From: Miroslav Lichvar [mailto:mli...@re...]
Sent: Friday, June 19, 2015 4:23 PM
To: Chandra Mallela
Cc: lin...@li...
Subject: Re: [Linuxptp-users] Interpretation of ptp4l results

On Fri, Jun 19, 2015 at 07:53:52AM +0000, Chandra Mallela wrote:
>
>
> Hi Friends,
>
> I would like to double-check whether my following interpretation is correct. Please elucidate me.
>
> For the following ptp4l output, the ptp4l reports rms value and the max deviation from the mean offset. Thus, if I have to report the worstcase accuracy, I take the (rms+max). For example, , looking at the first line, I see that the maximum offset uncertainty/inaccuracy  observed is 2 +7 = 9ns. The maximum frequency uncertainty is (156403 + 7) pbb.

The max value is just the maximum measured offset in that interval. It doesn't mean it's also the worst accuracy of the clock. There could be an asymmetry in the network delay and ptp4l wouldn't know the clock is actually behind or ahead of the time source. To evaluate the accuracy you would to measure the error independendly from PTP, for example by comparing PPS signal generated by the synchronized clock with the reference on a scope.

--
Miroslav Lichvar

________________________________

Confidentiality Notice.
This message may contain information that is confidential or otherwise protected from disclosure. If you are not the intended recipient, you are hereby notified that any use, disclosure, dissemination, distribution, or copying of this message, or any attachments, is strictly prohibited. If you have received this message in error, please advise the sender by reply e-mail, and delete the message and any attachments. Thank you.

Re: [Linuxptp-users] Interpretation of ptp4l results

[Chandra M. <sma...@al...>]

Hi Miroslav et al,

Perhaps I am not clear about my question - just would like to understand the mathematical interpretation of the following for the offset and frequency corrections. For example, in the following output,

Frequency correction: Average frequency deviation observed is 156403 ppb with the maximum deviation of 7 pbb.

Offset correction: the 'rms' (root mean squre) is the mean offset of all the offsets in the sample observed in the interval and 'max' is the maximum deviation observed from the rms value.


However, you mention that 'max' for the offset is simply the maximum offset observed in the sample. In that case, do frequency correction and offset correction values differ in their definitions (with 7ns in the offset defining the max offset in the sample whereas 7 pbb in the frequency showing the maximum deviation from the mean)?


ptp4l[833.369]: rms    2 max    7 freq -156403 +/-   7 delay    28 +/-   0

Thanking you in anticipation,
Regards,
Chandra

(c) : +60.175508142
(O): +60.4.636.6412

"Knowledge speaks, Wisdom listens"


-----Original Message-----
From: Chandra Mallela
Sent: Saturday, June 20, 2015 12:41 AM
To: 'Miroslav Lichvar'
Cc: lin...@li...
Subject: RE: [Linuxptp-users] Interpretation of ptp4l results



Hi Miroslav,

My question is about mathematical interpretation of the output from the ptp4l stack. If I see the output like the one below, is it fair to assume that the 'rms' (root mean squre) is the mean offset of all the sample offsets observed in the interval and 'max' is the maximum deviation observed from the rms value? I am likening it to kinda gaussian/normal distribution but without sign interpretation. Otherwise, I am fully in line with what you have mentioned below.

ptp4l[833.369]: rms    2 max    7 freq -156403 +/-   7 delay    28 +/-   0

Thanking you in anticipation,
Regards,
Chandra

(c) : +60.175508142
(O): +60.4.636.6412

"Knowledge speaks, Wisdom listens"



-----Original Message-----
From: Miroslav Lichvar [mailto:mli...@re...]
Sent: Friday, June 19, 2015 4:23 PM
To: Chandra Mallela
Cc: lin...@li...
Subject: Re: [Linuxptp-users] Interpretation of ptp4l results

On Fri, Jun 19, 2015 at 07:53:52AM +0000, Chandra Mallela wrote:
>
>
> Hi Friends,
>
> I would like to double-check whether my following interpretation is correct. Please elucidate me.
>
> For the following ptp4l output, the ptp4l reports rms value and the max deviation from the mean offset. Thus, if I have to report the worstcase accuracy, I take the (rms+max). For example, , looking at the first line, I see that the maximum offset uncertainty/inaccuracy  observed is 2 +7 = 9ns. The maximum frequency uncertainty is (156403 + 7) pbb.

The max value is just the maximum measured offset in that interval. It doesn't mean it's also the worst accuracy of the clock. There could be an asymmetry in the network delay and ptp4l wouldn't know the clock is actually behind or ahead of the time source. To evaluate the accuracy you would to measure the error independendly from PTP, for example by comparing PPS signal generated by the synchronized clock with the reference on a scope.

--
Miroslav Lichvar

________________________________

Confidentiality Notice.
This message may contain information that is confidential or otherwise protected from disclosure. If you are not the intended recipient, you are hereby notified that any use, disclosure, dissemination, distribution, or copying of this message, or any attachments, is strictly prohibited. If you have received this message in error, please advise the sender by reply e-mail, and delete the message and any attachments. Thank you.

Re: [Linuxptp-users] Interpretation of ptp4l results

[Richard C. <ric...@gm...>]

On Sun, Jun 21, 2015 at 11:09:08PM +0000, Chandra Mallela wrote:

> However, you mention that 'max' for the offset is simply the maximum
> offset observed in the sample. In that case, do frequency correction
> and offset correction values differ in their definitions (with 7ns
> in the offset defining the max offset in the sample whereas 7 pbb in
> the frequency showing the maximum deviation from the mean)?

These are two different parameters, and they are indeed defined
differently.  The offset is the observed offset based on the result of
the PTP.  The frequency is the output from the servo, and it is only
indirectly related to the observed offset.  For example, using a weak
proportional weight, you will have see a smaller range of frequencies.
Or if you prohibit resetting the clock, a large offset will cause a
constant frequency value (steering using the maximum possible value).

HTH,
Richard

Re: [Linuxptp-users] Interpretation of ptp4l results

[Chandra M. <sma...@al...>]

Hi Richard,

Sorry for belaboring on this.

Can I then confirm the following mathematical definitions of the ptp4l repor, based on our mail threadt? This helps all of us (mostly rookies like me) to be on clear mathematical understanding of what is reported.

When ptp4l reports:  ptp4l[833.369]: rms    2 max    7 freq -156403 +/-   7 delay    28 +/-   0


Offset correction: the 'rms' (root mean squre - 2 ns in the above case) is the mean offset (calculated as squareroot of the mean of the offset squares) of all the offsets in the sample observed in the interval and 'max' (7ns) is the maximum offset observed within the sample.
Frequency correction: Mean frequency deviation observed is 156403 ppb with the standard deviation of 7 pbb (if so, is 7pbb at 3sigma?): likening it to gaussian/normal distribution
Path delay: the mean value is 28ns and the standard deviation is 0: likening it to the gaussian/normal distribution

Thanking you in anticipation,
Regards,
Chandra

(c) : +60.175508142
(O): +60.4.636.6412

"Knowledge speaks, Wisdom listens"



-----Original Message-----
From: Richard Cochran [mailto:ric...@gm...]
Sent: Monday, June 22, 2015 2:31 PM
To: Chandra Mallela
Cc: Miroslav Lichvar; lin...@li...
Subject: Re: [Linuxptp-users] Interpretation of ptp4l results

On Sun, Jun 21, 2015 at 11:09:08PM +0000, Chandra Mallela wrote:

> However, you mention that 'max' for the offset is simply the maximum
> offset observed in the sample. In that case, do frequency correction
> and offset correction values differ in their definitions (with 7ns in
> the offset defining the max offset in the sample whereas 7 pbb in the
> frequency showing the maximum deviation from the mean)?

These are two different parameters, and they are indeed defined differently.  The offset is the observed offset based on the result of the PTP.  The frequency is the output from the servo, and it is only indirectly related to the observed offset.  For example, using a weak proportional weight, you will have see a smaller range of frequencies.
Or if you prohibit resetting the clock, a large offset will cause a constant frequency value (steering using the maximum possible value).

HTH,
Richard

________________________________

Confidentiality Notice.
This message may contain information that is confidential or otherwise protected from disclosure. If you are not the intended recipient, you are hereby notified that any use, disclosure, dissemination, distribution, or copying of this message, or any attachments, is strictly prohibited. If you have received this message in error, please advise the sender by reply e-mail, and delete the message and any attachments. Thank you.

Re: [Linuxptp-users] Interpretation of ptp4l results

[Richard C. <ric...@gm...>]

On Mon, Jun 22, 2015 at 12:22:01PM +0000, Chandra Mallela wrote:
> Offset correction: the 'rms' (root mean squre - 2 ns in the above
> case) is the mean offset (calculated as squareroot of the mean of
> the offset squares) of all the offsets in the sample observed in the
> interval and 'max' (7ns) is the maximum offset observed within the
> sample.

Yes.

> Frequency correction: Mean frequency deviation observed is 156403
> ppb with the standard deviation of 7 pbb

Yes.

> (if so, is 7pbb at 3sigma?):

?

> likening it to gaussian/normal distribution

??

> Path delay: the mean value is 28ns and the standard deviation is 0:

Yes.

> likening it to the gaussian/normal distribution

???

Take a look at stats.c to see how the stddev is being calculated.
It is really straightforward.

Thanks,
Richard

Re: [Linuxptp-users] Interpretation of ptp4l results

[Chandra M. <sma...@al...>]

Hi Richard,

Thank you very much. Yes, I need to look at stats.c block - trying to take advantage of your having gone through it already. I am sorry for that.

Thanking you in anticipation,
Regards,
Chandra

(c) : +60.175508142
(O): +60.4.636.6412

"Knowledge speaks, Wisdom listens"



-----Original Message-----
From: Richard Cochran [mailto:ric...@gm...]
Sent: Monday, June 22, 2015 9:13 PM
To: Chandra Mallela
Cc: Miroslav Lichvar; lin...@li...
Subject: Re: [Linuxptp-users] Interpretation of ptp4l results

On Mon, Jun 22, 2015 at 12:22:01PM +0000, Chandra Mallela wrote:
> Offset correction: the 'rms' (root mean squre - 2 ns in the above
> case) is the mean offset (calculated as squareroot of the mean of the
> offset squares) of all the offsets in the sample observed in the
> interval and 'max' (7ns) is the maximum offset observed within the
> sample.

Yes.

> Frequency correction: Mean frequency deviation observed is 156403 ppb
> with the standard deviation of 7 pbb

Yes.

> (if so, is 7pbb at 3sigma?):

?

> likening it to gaussian/normal distribution

??

> Path delay: the mean value is 28ns and the standard deviation is 0:

Yes.

> likening it to the gaussian/normal distribution

???

Take a look at stats.c to see how the stddev is being calculated.
It is really straightforward.

Thanks,
Richard

________________________________

Confidentiality Notice.
This message may contain information that is confidential or otherwise protected from disclosure. If you are not the intended recipient, you are hereby notified that any use, disclosure, dissemination, distribution, or copying of this message, or any attachments, is strictly prohibited. If you have received this message in error, please advise the sender by reply e-mail, and delete the message and any attachments. Thank you.