linuxptp-users

[Linuxptp-users] Question about meanPathDelay (possible bug?)

[Wolfgang W. <wol...@gm...>]

Hello,

I have a question about the calculation of the meanPathDelay by the  
functions process_delay_resp() and clock_path_delay().
The standard says the following about the meanPathDelay computation  
(section 11.3.2):
<meanPathDelay> shall be computed as: <meanPathDelay> = [(t2 - t3) +  
(receiveTimestamp of Delay_Resp message – originTimestamp of Sync message)  
– correctionField of Sync message – correctionField of Delay_Resp  
message]/2.

I'm not sure, but I think the "receiveTimestamp of Delay_Resp  
message"-part is not implemented correctly in LinuxPTP.

The function process_delay_resp() calls clock_path_delay() as follows:

	clock_path_delay(p->clock, p->delay_req->hwts.ts, m->ts.pdu,  
m->header.correction);

The third argument is used by clock_path_delay() as what is called t4 in  
the standard.
As far as I understand the code, the third argument (m->ts.pdu) is the  
ingress timestamp on the slave of the DelayResp-message.
But I think what should be passed here should be the ingress timestamp on  
the master of the DelayReq-message, which is contained in the  
"receiveTimestamp"-field of the DelayResp-message (rsp->receiveTimestamp).

Any comments on what is actually correct are welcome :)
I'm new to LinuxPTP and try to use it to improve my understanding of 1588.

regards,
Wolfgang Wallner

Re: [Linuxptp-users] Question about meanPathDelay (possible bug?)

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

On Tue, Jul 30, 2013 at 03:06:28PM +0200, Wolfgang Wallner wrote:
> I'm not sure, but I think the "receiveTimestamp of Delay_Resp  
> message"-part is not implemented correctly in LinuxPTP.
> 
> The function process_delay_resp() calls clock_path_delay() as follows:
> 
> 	clock_path_delay(p->clock, p->delay_req->hwts.ts, m->ts.pdu,  
> m->header.correction);
> 
> The third argument is used by clock_path_delay() as what is called t4 in  
> the standard.
> As far as I understand the code, the third argument (m->ts.pdu) is the  
> ingress timestamp on the slave of the DelayResp-message.
> But I think what should be passed here should be the ingress timestamp on  
> the master of the DelayReq-message, which is contained in the  
> "receiveTimestamp"-field of the DelayResp-message (rsp->receiveTimestamp).

To me, it seems it's the DelayResp receiveTimestamp in the ts.pdu
field and used as t4 in the delay calculation. It's filled by the
following code in the msg_post_recv() function.

        case DELAY_RESP:
		timestamp_post_recv(m, &m->delay_resp.receiveTimestamp);

Why do you think it's the ingress time of the DelayResp message?

-- 
Miroslav Lichvar

Re: [Linuxptp-users] Question about meanPathDelay (possible bug?)

[Wolfgang W. <wol...@gm...>]

On Tue, 30 Jul 2013 17:46:58 +0200, Miroslav Lichvar <mli...@re...>  
wrote:

> On Tue, Jul 30, 2013 at 03:06:28PM +0200, Wolfgang Wallner wrote:
>> I'm not sure, but I think the "receiveTimestamp of Delay_Resp
>> message"-part is not implemented correctly in LinuxPTP.
>>
>> The function process_delay_resp() calls clock_path_delay() as follows:
>>
>> 	clock_path_delay(p->clock, p->delay_req->hwts.ts, m->ts.pdu,
>> m->header.correction);
>>
>> The third argument is used by clock_path_delay() as what is called t4 in
>> the standard.
>> As far as I understand the code, the third argument (m->ts.pdu) is the
>> ingress timestamp on the slave of the DelayResp-message.
>> But I think what should be passed here should be the ingress timestamp  
>> on
>> the master of the DelayReq-message, which is contained in the
>> "receiveTimestamp"-field of the DelayResp-message  
>> (rsp->receiveTimestamp).
>
> To me, it seems it's the DelayResp receiveTimestamp in the ts.pdu
> field and used as t4 in the delay calculation. It's filled by the
> following code in the msg_post_recv() function.
>
>         case DELAY_RESP:
> 		timestamp_post_recv(m, &m->delay_resp.receiveTimestamp);

Oh, thanks for making this clear :)

>
> Why do you think it's the ingress time of the DelayResp message?

Because at first I misinterpreted the standard. When it says  
"receiveTimestamp of Delay_Resp message" I thought of the Ingress  
timestamp of DelayResp.
I was confused, and had a look at the source code of LinuxPTP. I found  
"m->ts.pdu", which at first looks like a timestamp assigned by a lower  
layer, e.g. the ingress timestamp.
I did not think of the possibility that the lower layer puts the value of  
delay_resp.receiveTimestamp into this field.

Sorry for broadcasting my confusion, I should have dug deeper in the code  
:)

Thanks for clearing this up,
Wolfgang

Re: [Linuxptp-users] Question about meanPathDelay (possible bug?)

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

On Wed, Jul 31, 2013 at 10:41:07AM +0200, Wolfgang Wallner wrote:
> I did not think of the possibility that the lower layer puts the value of  
> delay_resp.receiveTimestamp into this field.

pdu = protocol data unit, here protocol means the PTP.

There is a long comment in msg.h that describes the time stamp fields.

Thanks,
Richard

Re: [Linuxptp-users] Question about meanPathDelay (possible bug?)

[Wolfgang W. <wol...@gm...>]

On Wed, 31 Jul 2013 16:32:56 +0200, Richard Cochran  
<ric...@gm...> wrote:

> On Wed, Jul 31, 2013 at 10:41:07AM +0200, Wolfgang Wallner wrote:
>> I did not think of the possibility that the lower layer puts the value  
>> of
>> delay_resp.receiveTimestamp into this field.
>
> pdu = protocol data unit, here protocol means the PTP.
>
> There is a long comment in msg.h that describes the time stamp fields.

Do you mean the following comment?

	struct {
		/**
		 * Contains the time stamp from the packet data in a
		 * native binary format for the host machine. The
		 * exact source of the time stamp's value depends on
		 * the message type:
		 *
		 * - announce    originTimestamp
		 * - follow_up   preciseOriginTimestamp
		 * - sync        originTimestamp
		 * - delay_req   originTimestamp
		 * - pdelay_resp     requestReceiptTimestamp
		 * - pdelay_resp_fup responseOriginTimestamp
		 */
		struct timestamp pdu;
		/**
		 * Approximate ingress time stamp using the relative
		 * CLOCK_MONOTONIC. Used to determine when announce
		 * messages have expired.
		 */
		struct timespec host;
	} ts;

This comment does not fully match the behavior in the  
msg_post_recv()-function , as timestamps of delay_req-frames are not  
handled in this functions, and delay_resp-frames are missing in the  
comment description.
IMHO it would be more accurate to replace the delay_req-line with  
"delay_resp  receiveTimestamp".

regards, Wolfgang

Re: [Linuxptp-users] Question about meanPathDelay (possible bug?)

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

On Wed, Jul 31, 2013 at 05:30:17PM +0200, Wolfgang Wallner wrote:

> IMHO it would be more accurate to replace the delay_req-line with
> "delay_resp  receiveTimestamp".

Yes, you are right. Care to submit a patch?

Thanks,
Richard