Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
On 2013-06-20, Paul G bodo...@gmail.com wrote: On Sunday, May 26, 2013 5:25:28 PM UTC-4, unruh wrote: Lan will have delays of the order of .15ms (worse for gigabit) This not the delay reported by ntpq. Is it wrong or are you talking about another measurement? It is the time between the packet sent out and the packet received back, as reported in the peerstats file. \no idea or care what ntpq reports. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
Paul G wrote: I have four GPS receivers into four different computers. I've assumed that I should be able to get small offsets between them but to do so I need Basically, if this is important to you, you will need special hardware instrumentation. You will need a time standard that it known to be a lot better than any of the normal sources, and which generates something like a pulse per second output, probably a portable atomic clock. You will need an output pin on the PC that can be controlled directly from the PCI bus (no USB in between, even internally). You will need software on the PC that has direct access the I/O port controlling the pin (there must be no kernel code virtualising the device - you must have IO permissions at the hardware level. The software also needs to be able to read the interpolated time as quickly as possible, which probably means running in the kernel. Externally, you will need a gated counter and and a frequency source of period several times less than the desired measurement accuracy and accurate to better than one cycle over a second, possibly a 10MHz output from the atomic clock. Arrange the atomic clock to start the counter, on the second, and the output from the PC to stop it. Either side of setting the output from the PC read the interpolated software clock, and after the complete cycle, output the fractional seconds part of the time. Repeat this every second. You should then have enough information to calculate the actual PC software clock offset, from which you can decide which of your ntpd sources is best. Note, if you want to work to better than about 100ns, you will need to look up the data the data sheets for some of the logic used, to correct for the logic propagation delays. Alternatively you can just accept that there will always be some uncertainty in the measurements. to set time1. If that's the correct approach how do I pick the right system to be the benchmark. So far I've just been assuming that the purpose built M. Tharp server should have the lowest latency but I'm not sure -- in any case, at present, I can't adjust it. Each machine in the truncated table below (except 210) has a time1 value derived in an ad hoc fashion. l rrefid st delay offset jitter == 1 o22.0 .PPS.0 0.000 -0.001 0.001 2 +244 .PPS.1 0.0680.000 0.006 2 *210 .GPS.1 0.514 -0.008 0.136 2 +192 .PPS.1 0.4550.008 0.032 I feel as if this topic should be clearly explained somewhere but I've been unable to find it. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
On Thursday, June 20, 2013 5:51:50 PM UTC-4, E-Mail Sent to this address will be added to the BlackLists wrote: enable stats [etc. etc.] As noted earlier I've done that or it's not applicable. E.g. I only use the PPS driver and my seconds are numbered by an appliance that doesn't run ntpd. On one system at a time, ... have several other NTP servers configured {I usually shoot for 6 to 10} I don't have six to ten stratum one servers (but maybe I should) and it doesn't seem useful to compare my 500 microsecond offsets to non-local clocks. with all involved systems continuously running for more than one day; take an average ... In another thread (and some here) I explain how I've done that and I don't really like the e.g. 124 microsecond time1 I derived. However it does result in O(1) microsec. offsets between some of my clocks. perhaps the PPS signal is inverted? It's not. It seemed like David Taylor covered that on may 25th. Yes. While I appreciate the suggestions and the good will behind them they don't seem informed by my question/problem description. My key point is that ntpq appears to be telling me odd things. E.g. my network is low latency, symmetric and consistent but some of my offsets are one or two orders of magnitude beyond other offsets. So my question is how to find what I hope is hardware latency using the tools at hand or the coverse given multiple S1 clocks with O(10) microsecond offsets which one is right. I expect I will move a set of them to a 10/100 switch and see if that makes a difference. Ideally all my clocks pairs would look like this (both have time1 0): localhost oPPS(0) .PPS. ... 3770.0000.001 0.001 black +aster .PPS. ... 3770.065 -0.003 0.004 localhost oPPS(0) .PPS. ... 3770.0000.000 0.004 aster +black .PPS. ... 3770.0660.007 0.010 But maybe some of them are just not up to the task: aster *ntp1 .PPS. ... 3770.5260.129 0.166 ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
Paul G wrote: E.g. my network is low latency, symmetric and consistent but some of my offsets are one or two orders of magnitude beyond other offsets. Gigabit interrupt coalescing causes non deterministic asymmetric latency. I expect I will move a set of them to a 10/100 switch and see if that makes a difference. If it is due to gigabit interrupt coalescing it will make it more deterministic (although higher latency). -- E-Mail Sent to this address blackl...@anitech-systems.com will be added to the BlackLists. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
On Sunday, May 26, 2013 5:25:28 PM UTC-4, unruh wrote: Lan will have delays of the order of .15ms (worse for gigabit) This not the delay reported by ntpq. Is it wrong or are you talking about another measurement? ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
Paul G wrote: On Tuesday, June 4, 2013 3:44:19 PM UTC-4, E-Mail Sent to this address will be added to the BlackLists wrote: Is interrupt coalescing disabled on the gigabit ethernet adapters? Not by me. I assume when ntpq reports symmetric delays proportional to interface speed that interrupts are not the issue for my question. Perhaps that's a poor assumption. My question was -- given PPS clocks on the same network with ntpq offsets of: 0.021, 0.130, 0.064 and -0.046 -- any of which can be fudged away how do I determine the truth. That is how much is (consistent) hardware latency of any sort versus algorithmic restriction. e.g. delay, offset and jitter from two sides of a gigE connection: 0.073 -0.008 0.002 0.0640.011 0.006 Likewise fastE: 1.226 -0.059 0.076 1.1040.109 0.068 I don't really understand your question. When I have two pps sourced systems they will each be 10us offset but depending on network delay at that time ntpq from a different system will show offset of 100us. Ntpd tries to compensate for delays but best it can do is assume equal delays and maximum delay equal to total rtt delay. The interrupt delay is another variable of possibly few us unless there are queued interrupts. Loop_summary and peer_summary will produce daily averages but even then I see big differences (relating to temperature/load). David ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
Paul G wrote: BlackList wrote: Is interrupt coalescing disabled on the gigabit ethernet adapters? Not by me. I assume when ntpq reports symmetric delays proportional to interface speed that interrupts are not the issue for my question. Perhaps that's a poor assumption. Gigabit Ethernet adapter interrupt coalescing just makes the packet latency less deterministic, if you don't turn it off. -- E-Mail Sent to this address blackl...@anitech-systems.com will be added to the BlackLists. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
On 07/06/2013 19:02, unruh wrote: I did use gimp to label the axes and scales and to make the 5's look like 5's. I think tht the problem is that the font used for 6 pixel high text makes 5 look like 6. Have a look again to see if it is better. Much improved, Bill, thanks very much. -- Cheers, David Web: http://www.satsignal.eu ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
On 06/06/2013 20:17, Doug Calvert wrote: unruh has a nice page about interrupt coalescing: http://axion.physics.ubc.ca/scatter/rt.html This is a quite interesting note, unfortunately rendered rather less useful by the graphs, which have no axis names or units marked, and which appear to have inconsistent scales (although this may simply be lack of legibility). For example: http://axion.physics.ubc.ca/scatter/rtdt1.gif The units on the vertical axis appear to be labelled: -0.0001 -5x10-6 0 5x10-6 0.0001 which I interpret as -100us, -5 us, 0, +5us +100us, but it appears to be a linear axis. Is the 5x10-6 supposed to be 5x10-5? The character needs serious improvement as it looks like 5x10-6 to me. But in any case, as microseconds are referred to in the text, would it not be a lot clearer if the graphs were also in microseconds, rather than some small fraction of seconds? Similar problems in: http://axion.physics.ubc.ca/scatter/gps.gif where the -5 and -6 suffices are very difficult to distinguish. I do hope that Bill can re-do these graphs some time as it would enhance the value of this work. -- Cheers, David Web: http://www.satsignal.eu ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
On 2013-06-07, David Taylor david-tay...@blueyonder.co.uk.invalid wrote: On 06/06/2013 20:17, Doug Calvert wrote: unruh has a nice page about interrupt coalescing: http://axion.physics.ubc.ca/scatter/rt.html This is a quite interesting note, unfortunately rendered rather less useful by the graphs, which have no axis names or units marked, and which appear to have inconsistent scales (although this may simply be lack of legibility). For example: http://axion.physics.ubc.ca/scatter/rtdt1.gif The units on the vertical axis appear to be labelled: -0.0001 -5x10-6 0 5x10-6 0.0001 Yes, supermongo, the plotting program, does tend to make 5 and 6 similar at the small sizes used in superscripts. But as you say, it is a linear scale, and thus that must be a 5 rather than a 6. which I interpret as -100us, -5 us, 0, +5us +100us, but it appears to Since that is, as you point out, an interpretation that produces silly results, it is probably the interpretation that is at fault. be a linear axis. Is the 5x10-6 supposed to be 5x10-5? The character needs serious improvement as it looks like 5x10-6 to me. But in any case, as microseconds are referred to in the text, would it not be a lot clearer if the graphs were also in microseconds, rather than some small fraction of seconds? Similar problems in: http://axion.physics.ubc.ca/scatter/gps.gif where the -5 and -6 suffices are very difficult to distinguish. I do hope that Bill can re-do these graphs some time as it would enhance the value of this work. I am not at all sure that I can find the data anymore to redo the graphs. About all I could do is to use gimp to replace the axes labels and make the 5 look a bit more like 5 rather than 6. The problem is at the small scale of the superscripts, the little culicue up at the bottom of the 5 in the font used becomes a single pixel closing off the the loop, making it look a lot like a 6. So I used gimp to open it up again, and also to labels on the axes. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
On 07/06/2013 16:17, unruh wrote: On 2013-06-07, David Taylor david-tay...@blueyonder.co.uk.invalid wrote: On 06/06/2013 20:17, Doug Calvert wrote: unruh has a nice page about interrupt coalescing: http://axion.physics.ubc.ca/scatter/rt.html This is a quite interesting note, unfortunately rendered rather less useful by the graphs, which have no axis names or units marked, and which appear to have inconsistent scales (although this may simply be lack of legibility). For example: http://axion.physics.ubc.ca/scatter/rtdt1.gif The units on the vertical axis appear to be labelled: -0.0001 -5x10-6 0 5x10-6 0.0001 Yes, supermongo, the plotting program, does tend to make 5 and 6 similar at the small sizes used in superscripts. But as you say, it is a linear scale, and thus that must be a 5 rather than a 6. which I interpret as -100us, -5 us, 0, +5us +100us, but it appears to Since that is, as you point out, an interpretation that produces silly results, it is probably the interpretation that is at fault. be a linear axis. Is the 5x10-6 supposed to be 5x10-5? The character needs serious improvement as it looks like 5x10-6 to me. But in any case, as microseconds are referred to in the text, would it not be a lot clearer if the graphs were also in microseconds, rather than some small fraction of seconds? Similar problems in: http://axion.physics.ubc.ca/scatter/gps.gif where the -5 and -6 suffices are very difficult to distinguish. I do hope that Bill can re-do these graphs some time as it would enhance the value of this work. I am not at all sure that I can find the data anymore to redo the graphs. About all I could do is to use gimp to replace the axes labels and make the 5 look a bit more like 5 rather than 6. The problem is at the small scale of the superscripts, the little culicue up at the bottom of the 5 in the font used becomes a single pixel closing off the the loop, making it look a lot like a 6. So I used gimp to open it up again, and also to labels on the axes. Thanks, Bill. Sounds like an update to supermongo is needed! Can it label the axes? Not critical to regenerate the graphs, but perhaps if you add a little to the text to clarify scales and axes it may help others. Cheers, David -- Cheers, David Web: http://www.satsignal.eu ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
On 2013-06-07, David Taylor david-tay...@blueyonder.co.uk.invalid wrote: On 07/06/2013 16:17, unruh wrote: On 2013-06-07, David Taylor david-tay...@blueyonder.co.uk.invalid wrote: On 06/06/2013 20:17, Doug Calvert wrote: unruh has a nice page about interrupt coalescing: http://axion.physics.ubc.ca/scatter/rt.html This is a quite interesting note, unfortunately rendered rather less useful by the graphs, which have no axis names or units marked, and which appear to have inconsistent scales (although this may simply be lack of legibility). For example: http://axion.physics.ubc.ca/scatter/rtdt1.gif The units on the vertical axis appear to be labelled: -0.0001 -5x10-6 0 5x10-6 0.0001 Yes, supermongo, the plotting program, does tend to make 5 and 6 similar at the small sizes used in superscripts. But as you say, it is a linear scale, and thus that must be a 5 rather than a 6. which I interpret as -100us, -5 us, 0, +5us +100us, but it appears to Since that is, as you point out, an interpretation that produces silly results, it is probably the interpretation that is at fault. be a linear axis. Is the 5x10-6 supposed to be 5x10-5? The character needs serious improvement as it looks like 5x10-6 to me. But in any case, as microseconds are referred to in the text, would it not be a lot clearer if the graphs were also in microseconds, rather than some small fraction of seconds? Similar problems in: http://axion.physics.ubc.ca/scatter/gps.gif where the -5 and -6 suffices are very difficult to distinguish. I do hope that Bill can re-do these graphs some time as it would enhance the value of this work. I am not at all sure that I can find the data anymore to redo the graphs. About all I could do is to use gimp to replace the axes labels and make the 5 look a bit more like 5 rather than 6. The problem is at the small scale of the superscripts, the little culicue up at the bottom of the 5 in the font used becomes a single pixel closing off the the loop, making it look a lot like a 6. So I used gimp to open it up again, and also to labels on the axes. Thanks, Bill. Sounds like an update to supermongo is needed! Can it label the axes? Not critical to regenerate the graphs, but perhaps if you add a little to the text to clarify scales and axes it may help others. I did use gimp to label the axes and scales and to make the 5's look like 5's. I think tht the problem is that the font used for 6 pixel high text makes 5 look like 6. Have a look again to see if it is better. Cheers, David ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
unruh un...@invalid.ca wrote: I did use gimp to label the axes and scales and to make the 5's look like 5's. I think tht the problem is that the font used for 6 pixel high text makes 5 look like 6. Have a look again to see if it is better. At least one typo in the text: Finally it was changed to 1000 Mbps with rx-users: 0 (no interrupt coalescence. That should be rx-usecs. rick jones -- I don't interest myself in why. I think more often in terms of when, sometimes where; always how much. - Joubert these opinions are mine, all mine; HP might not want them anyway... :) feel free to post, OR email to rick.jones2 in hp.com but NOT BOTH... ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
On 2013-06-07, Rick Jones rick.jon...@hp.com wrote: unruh un...@invalid.ca wrote: I did use gimp to label the axes and scales and to make the 5's look like 5's. I think tht the problem is that the font used for 6 pixel high text makes 5 look like 6. Have a look again to see if it is better. At least one typo in the text: Finally it was changed to 1000 Mbps with rx-users: 0 (no interrupt coalescence. That should be rx-usecs. Thanks. Fixed. rick jones ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
unruh has a nice page about interrupt coalescing: http://axion.physics.ubc.ca/scatter/rt.html On Tue, Jun 4, 2013 at 3:44 PM, E-Mail Sent to this address will be added to the BlackLists Null@blacklist.anitech-systems.invalid wrote: Paul G wrote: Showing 192.168.0.2 and .244 with time1 = 0 as compared to the first post where time1 was ~100 microsec. These two have gigabit interfaces. Is interrupt coalescing disabled on the gigabit ethernet adapters? -- E-Mail Sent to this address blackl...@anitech-systems.com will be added to the BlackLists. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
Paul G wrote: Showing 192.168.0.2 and .244 with time1 = 0 as compared to the first post where time1 was ~100 microsec. These two have gigabit interfaces. Is interrupt coalescing disabled on the gigabit ethernet adapters? -- E-Mail Sent to this address blackl...@anitech-systems.com will be added to the BlackLists. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
On 27/05/2013 04:52, Paul G wrote: On Saturday, May 25, 2013 3:05:13 AM UTC-4, David Taylor wrote: What happens with the fudge set to zero? Please show the table for each machine. Showing 192.168.0.2 and .244 with time1 = 0 as compared to the first post where time1 was ~100 microsec. These two have gigabit interfaces. 0.2 is driven by a Garmin 18x and 0.244 is driven by a Firefly (I). 0.2 remote refid st t when poll reach delay offset jitter == o127.127.22.0.PPS.0 l18 3770.000 -0.001 0.002 +192.168.0.244 .PPS.1 u-8 3770.075 -0.128 0.001 *192.168.0.210 .GPS.1 u78 3770.5190.117 0.098 0.244 remote refid st t when poll reach delay offset jitter == o127.127.22.0.PPS.0 l58 3770.0000.000 0.002 +192.168.0.2 .PPS.1 u48 3770.0790.132 0.003 +192.168.0.210 .GPS.1 u48 3770.5470.253 0.135 *192.168.0.192 .PPS.1 u38 3770.4750.262 0.036 Well, that's interesting! Many thanks. 128/132 microseconds is way more than I would expect a GPS/PPS signal to be off - it should be nearer 1 microsecond (unless you're doing something unusual like a long line driver, or triggering off the wrong edge of a 130 microsecond pulse, perhaps?). I can't see a spec for the pulse width from the Firefly, but I do see it has a programmeable offset. I presume that's set to zero? Purely for comparison, on the systems here: - my FreeBSD/Intel Atom server sees my Windows XP system as -0.004, and my Win-7 system as -0.077. - my three Raspberry Pi cards see the FreeBSD server as -0.030 to +0.100. - my three Raspberry Pi cards see the Windows XP system as -0.040 to + 0.070, and my Win-7 system as -0.020 to + 0.130. One RasPi is connected via Wi-Fi. All of these are triggered from PPS signals which are well within 0.1 microseconds of each other when viewed on a digital storage scope. The 18x LVC signal to the FreeBSD node may be a little delayed because of the way it's being processed. I guess it says something about the internal delays in each PC, and perhaps something about the difference between network and serial/DCD processing. Each PC's clock is /not/ being set to UTC, but to some offset from UTC, which NTP measures as near zero. Cheers, David -- Cheers, David Web: http://www.satsignal.eu ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
Hi Paul! On Fri, May 24, 2013 at 9:28 PM, Paul G bodo...@gmail.com wrote: I have four GPS receivers into four different computers. I've assumed that I should be able to get small offsets between them but to do so I need to set time1. If that's the correct approach how do I pick the right system to be the benchmark. So far I've just been assuming that the purpose built M. Tharp server should have the lowest latency but I'm not sure -- in any case, at present, I can't adjust it. Each machine in the truncated table below (except 210) has a time1 value derived in an ad hoc fashion. l rrefid st delay offset jitter == 1 o22.0 .PPS.0 0.000 -0.001 0.001 2 +244 .PPS.1 0.0680.000 0.006 2 *210 .GPS.1 0.514 -0.008 0.136 2 +192 .PPS.1 0.4550.008 0.032 I feel as if this topic should be clearly explained somewhere but I've been unable to find it. My recent experience with 4 stratum 1 servers (one of them an appliance from Meinberg - great box BTW) is that you will never have every server in sync with each other due to the latency and delays of the network (even LAN, same segment as is in this case). I don't depend on external NTP servers. My Meinberg NTP server: $ ntpq -p 10.0.2.1 remote refid st t when poll reach delay offset jitter == LOCAL(0).LOCL. 10 l8 16 3770.0000.000 0.001 +GENERIC(0) .GPS.0 l 13 16 3770.0000.000 0.001 oPPS(0) .PPS.0 l7 16 3770.000 -0.001 0.001 -10.0.2.2.PPS.1 u 37 64 3770.1990.007 0.031 -10.0.2.3.PPS.1 u6 64 3770.171 -0.004 0.008 +10.0.2.4.PPS.1 u 51 64 3770.157 -0.002 0.009 My Garmin 18 LVC on the roof: $ ntpq -p 10.0.2.2 remote refid st t when poll reach delay offset jitter == oGPS_NMEA(0) .PPS.0 l5 16 3770.0000.000 0.004 +10.0.2.1.PPS.1 u2 16 3770.1570.005 0.013 +10.0.2.3.PPS.1 u1 16 3770.144 -0.001 0.004 +10.0.2.4.PPS.1 u 16 16 3770.140 -0.002 0.004 My Sure GPS board on the window sill: $ ntpq -p 10.0.2.3 remote refid st t when poll reach delay offset jitter == oGPS_NMEA(0) .PPS.0 l 13 16 3770.0000.001 0.004 +10.0.2.1.PPS.1 u 10 16 3770.1660.003 0.008 +10.0.2.2.PPS.1 u5 16 3770.1380.004 0.004 +10.0.2.4.PPS.1 u8 16 3770.1400.001 0.004 My 2nd Sure GPS board on the same window sill: $ ntpq -p 10.0.2.4 remote refid st t when poll reach delay offset jitter == oGPS_NMEA(0) .PPS.0 l7 16 3770.0000.000 0.004 +10.0.2.1.PPS.1 u4 16 3770.2010.000 0.011 +10.0.2.2.PPS.1 u3 16 3770.1360.002 0.004 +10.0.2.3.PPS.1 u2 16 3770.1400.001 0.004 For me 10 us offsets is OK and acceptable between stratum 1 machines. I believe this is what is expected. Regards, Miguel ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
On 2013-05-26, Miguel Gon?alves miguel.barbosa.goncal...@gmail.com wrote: Hi Paul! On Fri, May 24, 2013 at 9:28 PM, Paul G bodo...@gmail.com wrote: I have four GPS receivers into four different computers. I've assumed that I should be able to get small offsets between them but to do so I need to set time1. If that's the correct approach how do I pick the right system to be the benchmark. So far I've just been assuming that the purpose built M. Tharp server should have the lowest latency but I'm not sure -- in any case, at present, I can't adjust it. Each machine in the truncated table below (except 210) has a time1 value derived in an ad hoc fashion. l rrefid st delay offset jitter == 1 o22.0 .PPS.0 0.000 -0.001 0.001 2 +244 .PPS.1 0.0680.000 0.006 2 *210 .GPS.1 0.514 -0.008 0.136 2 +192 .PPS.1 0.4550.008 0.032 I feel as if this topic should be clearly explained somewhere but I've been unable to find it. My recent experience with 4 stratum 1 servers (one of them an appliance from Meinberg - great box BTW) is that you will never have every server in sync with each other due to the latency and delays of the network (even LAN, same segment as is in this case). I don't depend on external NTP servers. Lan will have delays of the order of .15ms (worse for gigabit) with accuracy of the order of 10-20us. Direct PPS will be of order 2-5us, but only if you have only one. If you have more than one, then they get in the way of each other ( the interrupts all arrive at the same time, and each has to wait for th other to finish which is about 10us. ) My Meinberg NTP server: $ ntpq -p 10.0.2.1 remote refid st t when poll reach delay offset jitter == LOCAL(0).LOCL. 10 l8 16 3770.0000.000 0.001 +GENERIC(0) .GPS.0 l 13 16 3770.0000.000 0.001 oPPS(0) .PPS.0 l7 16 3770.000 -0.001 0.001 -10.0.2.2.PPS.1 u 37 64 3770.1990.007 0.031 -10.0.2.3.PPS.1 u6 64 3770.171 -0.004 0.008 +10.0.2.4.PPS.1 u 51 64 3770.157 -0.002 0.009 My Garmin 18 LVC on the roof: $ ntpq -p 10.0.2.2 remote refid st t when poll reach delay offset jitter == oGPS_NMEA(0) .PPS.0 l5 16 3770.0000.000 0.004 +10.0.2.1.PPS.1 u2 16 3770.1570.005 0.013 +10.0.2.3.PPS.1 u1 16 3770.144 -0.001 0.004 +10.0.2.4.PPS.1 u 16 16 3770.140 -0.002 0.004 My Sure GPS board on the window sill: $ ntpq -p 10.0.2.3 remote refid st t when poll reach delay offset jitter == oGPS_NMEA(0) .PPS.0 l 13 16 3770.0000.001 0.004 +10.0.2.1.PPS.1 u 10 16 3770.1660.003 0.008 +10.0.2.2.PPS.1 u5 16 3770.1380.004 0.004 +10.0.2.4.PPS.1 u8 16 3770.1400.001 0.004 My 2nd Sure GPS board on the same window sill: $ ntpq -p 10.0.2.4 remote refid st t when poll reach delay offset jitter == oGPS_NMEA(0) .PPS.0 l7 16 3770.0000.000 0.004 +10.0.2.1.PPS.1 u4 16 3770.2010.000 0.011 +10.0.2.2.PPS.1 u3 16 3770.1360.002 0.004 +10.0.2.3.PPS.1 u2 16 3770.1400.001 0.004 For me 10 us offsets is OK and acceptable between stratum 1 machines. I believe this is what is expected. Regards, Miguel ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
On Saturday, May 25, 2013 1:06:48 AM UTC-4, unruh wrote: GPS receivers are not particularly good time sources UNLESS you use PPS. All of my clocks use PPS and in fact none of them use GPS to number the seconds except the one that uses refid GPS. That clock is a purpose built NTP-equivalent server called a Laureline which uses PPS and numbers seconds via serial input. Per the refclock document the PPS/NMEA (20/22) drivers use time1 to correct for PPS offset while 22 uses time2 to correct for serial latency. Since the clocks differ by O(100 microsec) without time1 correction I assumed that was internal rather than network latency. If you use PPS they should all agree to the 1micro second level. They do per ntpq given resp. 124, 151 and 85 microsecond time1. grabbing times via the net from the those computers will be dominated by network issues. I've been assuming that since I get fairly consistent offsets (fudged or not) and jitter that network latency is not the culprit. But I'm clearly confused about something. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
On Saturday, May 25, 2013 3:05:13 AM UTC-4, David Taylor wrote: What happens with the fudge set to zero? Please show the table for each machine. Showing 192.168.0.2 and .244 with time1 = 0 as compared to the first post where time1 was ~100 microsec. These two have gigabit interfaces. 0.2 is driven by a Garmin 18x and 0.244 is driven by a Firefly (I). 0.2 remote refid st t when poll reach delay offset jitter == o127.127.22.0.PPS.0 l18 3770.000 -0.001 0.002 +192.168.0.244 .PPS.1 u-8 3770.075 -0.128 0.001 *192.168.0.210 .GPS.1 u78 3770.5190.117 0.098 0.244 remote refid st t when poll reach delay offset jitter == o127.127.22.0.PPS.0 l58 3770.0000.000 0.002 +192.168.0.2 .PPS.1 u48 3770.0790.132 0.003 +192.168.0.210 .GPS.1 u48 3770.5470.253 0.135 *192.168.0.192 .PPS.1 u38 3770.4750.262 0.036 ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
On 24/05/2013 21:28, Paul G wrote: I have four GPS receivers into four different computers. I've assumed that I should be able to get small offsets between them but to do so I need to set time1. If that's the correct approach how do I pick the right system to be the benchmark. So far I've just been assuming that the purpose built M. Tharp server should have the lowest latency but I'm not sure -- in any case, at present, I can't adjust it. Each machine in the truncated table below (except 210) has a time1 value derived in an ad hoc fashion. l rrefid st delay offset jitter == 1 o22.0 .PPS.0 0.000 -0.001 0.001 2 +244 .PPS.1 0.0680.000 0.006 2 *210 .GPS.1 0.514 -0.008 0.136 2 +192 .PPS.1 0.4550.008 0.032 I feel as if this topic should be clearly explained somewhere but I've been unable to find it. Paul, Those PCs with PPS feed should have zero offset, and time1 should be set to zero unless you /know/ that the GPS PPS signal is offset (e.g. you are feeding it via several microseconds delay of cable, which is unlikely). What NTP is actually syncing to is an internal PPS (if you like), i.e. the external PPS signal delayed by whatever processing is taking place inside the PC, but any external things you measure will be subject to the same delay. What happens with the fudge set to zero? Please show the table for each machine. For the PC with just GPS sync, you might get it a PPS feed somehow, use PPS as the reference, set GPS/NMEA to noselect in the ntp.conf, and record the peersstats. I wrote this up with help from others here: http://www.satsignal.eu/ntp/Garmin-GSP18x-LVC-firmware-issue.htm#analysis and although the results in that case are wrong, the method is, I believe, correct. -- Cheers, David Web: http://www.satsignal.eu ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
[ntp:questions] How to determin hardware latency for PPS offset given simple tools.
I have four GPS receivers into four different computers. I've assumed that I should be able to get small offsets between them but to do so I need to set time1. If that's the correct approach how do I pick the right system to be the benchmark. So far I've just been assuming that the purpose built M. Tharp server should have the lowest latency but I'm not sure -- in any case, at present, I can't adjust it. Each machine in the truncated table below (except 210) has a time1 value derived in an ad hoc fashion. l rrefid st delay offset jitter == 1 o22.0 .PPS.0 0.000 -0.001 0.001 2 +244 .PPS.1 0.0680.000 0.006 2 *210 .GPS.1 0.514 -0.008 0.136 2 +192 .PPS.1 0.4550.008 0.032 I feel as if this topic should be clearly explained somewhere but I've been unable to find it. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] How to determin hardware latency for PPS offset given simple tools.
On 2013-05-24, Paul G bodo...@gmail.com wrote: I have four GPS receivers into four different computers. I've assumed that I GPS receivers are not particularly good time sources UNLESS you use PPS. If you use PPS they should all agree to the 1micro second level. You should NOT need time1. If youjust use the serial data, you will be lucky to get millisecond agreement, and that only with a lot of work. Ie, if you are going to set up gps, set up pps. Since you seem to be using it, you should get us accuracy from them without any offset. grabbing times via the net from the those computers will be dominated by network issues. should be able to get small offsets between them but to do so I need to set time1. If that's the correct approach how do I pick the right system to be the benchmark. So far I've just been assuming that the purpose built M. Tharp server should have the lowest latency but I'm not sure -- in any case, at present, I can't adjust it. Each machine in the truncated table below (except 210) has a time1 value derived in an ad hoc fashion. Are they a secret? l rrefid st delay offset jitter == 1 o22.0 .PPS.0 0.000 -0.001 0.001 2 +244 .PPS.1 0.0680.000 0.006 2 *210 .GPS.1 0.514 -0.008 0.136 2 +192 .PPS.1 0.4550.008 0.032 I feel as if this topic should be clearly explained somewhere but I've been unable to find it. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions