Hello linuxptp-devel group!
I would like to discuss the implementation of the two-step peer delay
computation. The 1588 and 802.1AS specifications unfortunately have a small
discrepancy in their use of correction fields for peer delay. This project
appears to be using the 1588 equation rather than the 802.1AS equation.
This is generally not an issue, since 802.1AS implementations typically set
this field to zero, however, there is at least one switch IC on the market,
the KSZ9567, with a silicon errata that results in a non-zero value in this
field. The current version of the MOTU switch uses this IC and implements a
workaround based on the 802.1AS equation. Since the two equations differ,
there is an interoperability issue with the linuxptp project and the MOTU
switch.
The 1588 specification [section 11.4.3 d) 4) on p.116 of the IEEE
1588-2008] specifies the following equation to be used to compute mean path
delay:
<meanPathDelay> = [
(t4 - t1) -
(responseOriginTimestamp - requestReceiptTimestamp) -
correctionField of Pdelay_Resp -
correctionField of Pdelay_Resp_Follow_Up
]/2
where:
t4 = Pdelay_Resp Ingress Timestamp
t1 = Pdelay_Req Egress Timestamp
The 802.1AS specification [section 11.2.19.3.4 computePropTime() on p.186
of 802.1AS-2020] specifies the following equation:
D = (r * (t4 - t1) - (t3 - t2)) / 2
where:
t4 = pdelayRespEventIngressTimestamp
t1 = pdelayReqEventEgressTimestamp
t2 = sum of
1) The nanoseconds field of the requestReceiptTimestamp,
2) The seconds field of the requestReceiptTimestamp, multiplied by
10^9 , and
3) The correction field, divided by 2^16, of the Pdelay_Resp message
t3 = sum of
1) The nanoseconds field of the responseOriginTimestamp,
2) The seconds field of the responseOriginTimestamp, multiplied by
10^9 , and
3) The correction field, divided by 2^16, of the Pdelay_Resp_Follow_Up
message
r = current value of neighborRateRatio
In short, the difference is that the 1588 specification asks the
implementation to sum the correction fields whereas the 802.1AS
specification asks the implementation to take their difference. The 802.1AS
specification seems to almost go as far as saying that the peer delay
correction fields can be ignored altogether (see the NOTE in table 11-6 on
p.199 of 802.1AS-2020), and I am aware of a few implementations that do
ignore it without apparent issue. I am interested in making the linuxptp
project compatible with MOTU switches based on the KSZ9567 switch IC, but
before submitting any patches I wanted to explain the motivation and see if
this is something that the project would be open to considering.
--
Dylan Robinson
Software Engineer
MOTU
_______________________________________________
Linuxptp-devel mailing list
Linuxptp-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linuxptp-devel
