> So first of all PTP doesn't compensate for anything at all is that right?
Linuxptp has a delayAsymmetry configuration parameter, see the man page and source code for details. From memory the magnitude of the asymmetry of the delay between the two endpoints is what skews the offset calculation. (Richard/Someone please confirm this) However, I would try to address the delay issue first and then if still necessary use the delayAsymmetry configuration parameter. I would connect two i210 point to point (directly), run linuxptp and change the egress/ingress compensation values until: 1- The peer delay is symmetric 2- The peer delay is close to what you expect for the length of the cable you are using to connect the two peers (short patch cables are not a good idea for this case, I would use a cable a few meters long) I would then repeat the above procedure for two of your FEC based devices connected point to point. Then I would go back to your i210 <—> FEC setup and check the peer delay and PPS results again using the compensation values I found in the previous steps. > (Is there a way todo this automatically?) No, you would need another mechanism to measure the asymmetry and that it’s outside the scope of IEE1588 AFAIK. Also, at gigabit link speed it is common for delays to have quite a bit of variance (a few 100s of ns in my experience). HTH — Delio > On 5 Jul 2019, at 07:45, Armin HAMAR <armin.ha...@sprecher-automation.com> > wrote: > > So first of all PTP doesn't compensate for anything at all is that right? (Is > there a way todo this automatically?) > Found out that the delay i210 with another i210 is pretty consistent (very > few nanoseconds). > It cancels only out if i have two completly identical card's? > > That the intel driver speaks not a hundred percent proof is definitfly a > thing. > Now i would need to proof which of them produces more in/egress latency. > > > > Von: Miroslav Lichvar <mlich...@redhat.com> > An: Richard Cochran <richardcoch...@gmail.com> > Kopie: Armin HAMAR <armin.ha...@sprecher-automation.com>, > Linuxptp-users@lists.sourceforge.net > Datum: 03.07.2019 08:06 > Betreff: Re: [Linuxptp-users] PPS to Linux PTP offset/delay? > > > > On Tue, Jul 02, 2019 at 09:31:44AM -0700, Richard Cochran wrote: > > On Tue, Jul 02, 2019 at 09:36:50AM +0200, Miroslav Lichvar wrote: > > > So, with two > > > directly connected I210s (short cable) the measured delay should be > > > close to zero. > > > > That is true of any two devices, since the tx/rx asymmetry cancels out > > (even without any corrections applied). > > The error in the measured offset cancels out between any two identical > NICs, but the measured PTP delay is a sum of the TX and RX > timestamping errors (and the delay in the cable). > > For example between two I350s on 1Gb I see a delay of 507 ns. An > offset I measured between an I210 (using the igb compensation) and > I350 in the same computer over PCIe is about 240 ns, so the TX/RX > error of the I350 seems to be very asymmetric too. > > -- > Miroslav Lichvar > > > _______________________________________________ > Linuxptp-users mailing list > Linuxptp-users@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/linuxptp-users _______________________________________________ Linuxptp-users mailing list Linuxptp-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/linuxptp-users
