Re: [ntp:questions] NTP server with Garmin GPS 18 LVC
On 18/12/2013 19:59, Adrian P wrote: [] Thanks for the great site! I just noticed that mine is actually a GPS 18x LVC, as written on its back. It is temporary located on the inside window sill, in my apartment at the 3rd floor. So it can see only half of the sky, through the window. Later I plan to install it on the roof, but first I need to figure out how to route the cable outside of my apartment (need to drill one hole I suppose :). From your graphs I could see that the best time keeper is the Raspberry Pi #1, running Linux/3.6.11 and the Adafruit MTK3339 indoors RX. Do I read them correctly? Everywhere is written that FreeBSD is the best OS for time keeper, since its kernel supports nanosecond accuracies. However, if GPS 18x LVC PPS signal has only microsecond accuracy, then I suppose there is no point on using FreeBSD, any linux would be okay as well. So what do you think, looking from NTP point of view only, is this nanosecond accuracy the only reason why someone would use FreeBSD, or there are some other reasons as well? I see you have experience with all kind of OSes. Many thanks, Adrian Glad you like the site, Adrian! Seeing half the sky is likely good enough for the 18x, but do beware of windows - some have a thin metallic coating to reduce IR incoming, but that can also reduce 1.57 GHz RF incoming as well! Try with your mobile phone to see whether the better GPS signals are though the window or behind the wall! There are GPS status programs for iPhone and Android. The Raspberry Pi cards are all good timekeepers, largely independent of the GPS used (providing it offers PPS output). The accuracy of the PPS is about one order of magnitude better than any of today's PCs can manage, so it's really a choice between soldering or not: Soldering and has on-board antenna: http://www.adafruit.com/products/746 No soldering puck antenna: http://ava.upuaut.net/store/index.php?route=product/productpath=59_60product_id=95 I haven't investigated why, from the graphs, Raspi-3 and Raspi-4 appear to be slightly worse. Raspi-3 is also in active use as an ADS-B receiver and is sending data continually to two PC clients which process and display the data. Raspi-4 runs a digital wall clock, but that doesn't take a lot of CPU or I/O, but it does mean that it's running a graphical display whereas all the others are SSH access. I don't believe that the poorer performance is because of the particular GPS used. In the past, FreeBSD was certainly reported to be better than Linux, but I don't believe that to be the case today. Purely for time-keeping I see little or no difference. Whether Free BSD has fewer variants than Linux, and whether that might make it easier to support may be considerations, but I'm not really competent to comment on those. What accuracy do you need? I'm after millisecond level on Windows PCs, which can be achieved with a PPS source, and should be even easier to achieve in Windows 8.1 (compared to 7 or earlier). To sync a PC to the full accuracy which a time-keeping GPS can deliver would take considerable effort. It would be interesting to know whether anyone has actually achieved that, and if so, how! -- Cheers, David Web: http://www.satsignal.eu ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] NTP server with Garmin GPS 18 LVC
Adrian P wrote: . Everywhere is written that FreeBSD is the best OS for time keeper, since its kernel supports nanosecond accuracies. However, if GPS 18x LVC PPS signal has only microsecond accuracy, then I suppose there is no point on using FreeBSD, any linux would be okay as well. So what do you think, looking from NTP point of view only, is this nanosecond accuracy the only reason why someone would use FreeBSD, or there are some other reasons as well? I see you have experience with all kind of OSes. I think the combination of hardware used is at least as important as the OS. http://www.febo.com/pages/soekris/ 100ns accuracy http://www.synclab.org/radclock/ An alternative to ntpd David ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
[ntp:questions] 4.2.4p8 - 4.2.6p5 less frequent driftfile updates?
Hello everyone, about 40h ago, I updated half a platform from CentOS 6.4 to 6.5, which includes updating ntpd from ntp-4.2.4p8-3.el6.centos.x86_64 to ntp-4.2.6p5-1.el6.centos.x86_64. One of the updated machines holds the platform's Internet-bound NTP connections, and has some extra monitoring for that - which now alerts us to the fact that by now, the driftfile gets updated only every 6+ hours, rather than every hour (modulo sync problems) as before. A quick scan seems to confirm that the update frequency is tied to the ntpd version running on the machines: $ date ; for MACHINE in $INTERNETFACING ; do echo ; ssh $MACHINE \ rpm -q ntp ; ls -l /var/lib/ntp/drift 2/dev/null ; done Mi 18. Dez 19:03:06 CET 2013 ntp-4.2.6p5-1.el6.centos.x86_64 -rw-r--r--. 1 ntp ntp 7 18. Dez 15:18 /var/lib/ntp/drift ntp-4.2.6p5-1.el6.centos.x86_64 -rw-r--r--. 1 ntp ntp 7 18. Dez 18:20 /var/lib/ntp/drift ntp-4.2.4p8-3.el6.centos.x86_64 -rw-r--r--. 1 ntp ntp 7 18. Dez 18:11 /var/lib/ntp/drift ntp-4.2.4p8-3.el6.centos.x86_64 -rw-r--r--. 1 ntp ntp 7 18. Dez 18:29 /var/lib/ntp/drift ntp-4.2.4p8-3.el6.centos.x86_64 -rw-r--r--. 1 ntp ntp 7 18. Dez 18:21 /var/lib/ntp/drift ntp-4.2.6p5-1.el6.centos.x86_64 -rw-r--r--. 1 ntp ntp 7 18. Dez 14:13 /var/lib/ntp/drift ntp-4.2.6p5-1.el6.centos.x86_64 -rw-r--r--. 1 ntp ntp 7 18. Dez 13:22 /var/lib/ntp/drift Is this change of behavior intentional? Kind regards, J. Bern -- *NEU* - NEC IT-Infrastruktur-Produkte im http://www.linworks-shop.de/: Server--Storage--Virtualisierung--Management SW--Passion for Performance Jochen Bern, Systemingenieur --- LINworks GmbH http://www.LINworks.de/ Postfach 100121, 64201 Darmstadt | Robert-Koch-Str. 9, 64331 Weiterstadt PGP (1024D/4096g) FP = D18B 41B1 16C0 11BA 7F8C DCF7 E1D5 FAF4 444E 1C27 Tel. +49 6151 9067-231, Zentr. -0, Fax -299 - Amtsg. Darmstadt HRB 85202 Unternehmenssitz Weiterstadt, Geschäftsführer Metin Dogan, Oliver Michel ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] 4.2.4p8 - 4.2.6p5 less frequent driftfile updates?
Jochen Bern wrote: Hello everyone, about 40h ago, I updated half a platform from CentOS 6.4 to 6.5, which includes updating ntpd from ntp-4.2.4p8-3.el6.centos.x86_64 to ntp-4.2.6p5-1.el6.centos.x86_64. One of the updated machines holds the platform's Internet-bound NTP connections, and has some extra monitoring for that - which now alerts us to the fact that by now, the driftfile gets updated only every 6+ hours, rather than every hour (modulo sync problems) as before. A quick scan seems to confirm that the update frequency is tied to the ntpd version running on the machines: This is intentional, in order to lace less load on the hard drive/flash drive of embedded/small servers. :-) The update is written only if the delta in time or frequency is large enough afair. Terje $ date ; for MACHINE in $INTERNETFACING ; do echo ; ssh $MACHINE \ rpm -q ntp ; ls -l /var/lib/ntp/drift 2/dev/null ; done Mi 18. Dez 19:03:06 CET 2013 ntp-4.2.6p5-1.el6.centos.x86_64 -rw-r--r--. 1 ntp ntp 7 18. Dez 15:18 /var/lib/ntp/drift ntp-4.2.6p5-1.el6.centos.x86_64 -rw-r--r--. 1 ntp ntp 7 18. Dez 18:20 /var/lib/ntp/drift ntp-4.2.4p8-3.el6.centos.x86_64 -rw-r--r--. 1 ntp ntp 7 18. Dez 18:11 /var/lib/ntp/drift ntp-4.2.4p8-3.el6.centos.x86_64 -rw-r--r--. 1 ntp ntp 7 18. Dez 18:29 /var/lib/ntp/drift ntp-4.2.4p8-3.el6.centos.x86_64 -rw-r--r--. 1 ntp ntp 7 18. Dez 18:21 /var/lib/ntp/drift ntp-4.2.6p5-1.el6.centos.x86_64 -rw-r--r--. 1 ntp ntp 7 18. Dez 14:13 /var/lib/ntp/drift ntp-4.2.6p5-1.el6.centos.x86_64 -rw-r--r--. 1 ntp ntp 7 18. Dez 13:22 /var/lib/ntp/drift Is this change of behavior intentional? Kind regards, J. Bern -- - Terje.Mathisen at tmsw.no almost all programming can be viewed as an exercise in caching ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] NTP server with Garmin GPS 18 LVC
On 2013-12-19, David Taylor david-tay...@blueyonder.co.uk.invalid wrote: On 18/12/2013 19:59, Adrian P wrote: ... What accuracy do you need? I'm after millisecond level on Windows PCs, which can be achieved with a PPS source, and should be even easier to achieve in Windows 8.1 (compared to 7 or earlier). To sync a PC to the full accuracy which a time-keeping GPS can deliver would take considerable effort. It would be interesting to know whether anyone has actually achieved that, and if so, how! Certainly Linux and FreeBSD are capable of microsecond, not millisecond, accuracy on modern equipment. For better than that, I think you would need special interrupt or timestamping hardware to get around the interrupt latency and variability of linux. While averaging the fluctuations, as chrony does, could deliver sub microsecond uncertainty, it is not clear you can get that accuracy due to average latency (Ie, if the timestamp on the interrupts is on average 1us late, all the averaging in the world will not get rid of that.-- That was the kind of figures I got when I timestamped a parallel port output pin transition and timestamped the parallel port interrupt that that pin was tied to-- 1-2us latency. Now that was many years ago, but I doubt that things have gotten better-- no reason for them to get better. Noone else really wants that kind of low latency interrupts, and interrupt conflicts cause much more variation than that. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] NTP server with Garmin GPS 18 LVC
On 2013-12-19, David Lord sn...@lordynet.org wrote: Adrian P wrote: . Everywhere is written that FreeBSD is the best OS for time keeper, since its kernel supports nanosecond accuracies. However, if GPS 18x LVC PPS signal has only microsecond accuracy, then I suppose there is no point on using FreeBSD, any linux would be okay as well. So what do you think, looking from NTP point of view only, is this nanosecond accuracy the only reason why someone would use FreeBSD, or there are some other reasons as well? I see you have experience with all kind of OSes. I think the combination of hardware used is at least as important as the OS. http://www.febo.com/pages/soekris/ 100ns accuracy AGreed. But their comparison is strongly cooked. At least on my system, the serial port interrupt performace is far better than theirs-- something like 2 us, not the 50us they are showing. Certainly 100ns is better, but more like one order of magnitude, not more than 2. http://www.synclab.org/radclock/ An alternative to ntpd David ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] NTP server with Garmin GPS 18 LVC
unruh wrote: On 2013-12-19, David Lord sn...@lordynet.org wrote: Adrian P wrote: . Everywhere is written that FreeBSD is the best OS for time keeper, since its kernel supports nanosecond accuracies. However, if GPS 18x LVC PPS signal has only microsecond accuracy, then I suppose there is no point on using FreeBSD, any linux would be okay as well. So what do you think, looking from NTP point of view only, is this nanosecond accuracy the only reason why someone would use FreeBSD, or there are some other reasons as well? I see you have experience with all kind of OSes. I think the combination of hardware used is at least as important as the OS. http://www.febo.com/pages/soekris/ 100ns accuracy AGreed. But their comparison is strongly cooked. At least on my system, the serial port interrupt performace is far better than theirs-- something like 2 us, not the 50us they are showing. Certainly 100ns is better, but more like one order of magnitude, not more than 2. It might not be very clear but PPS to serial port isn't used other than as a comparison, ntpd has been modified to use the Elan Timer with PPS input to the GPIO. I have a couple of motherboards with chipsets that have higher resolution timestamping but the GPIO pins aren't accessible. David http://www.synclab.org/radclock/ An alternative to ntpd David ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] NTP server with Garmin GPS 18 LVC
On 2013-12-19, David Lord sn...@lordynet.org wrote: unruh wrote: On 2013-12-19, David Lord sn...@lordynet.org wrote: Adrian P wrote: . Everywhere is written that FreeBSD is the best OS for time keeper, since its kernel supports nanosecond accuracies. However, if GPS 18x LVC PPS signal has only microsecond accuracy, then I suppose there is no point on using FreeBSD, any linux would be okay as well. So what do you think, looking from NTP point of view only, is this nanosecond accuracy the only reason why someone would use FreeBSD, or there are some other reasons as well? I see you have experience with all kind of OSes. I think the combination of hardware used is at least as important as the OS. http://www.febo.com/pages/soekris/ 100ns accuracy AGreed. But their comparison is strongly cooked. At least on my system, the serial port interrupt performace is far better than theirs-- something like 2 us, not the 50us they are showing. Certainly 100ns is better, but more like one order of magnitude, not more than 2. It might not be very clear but PPS to serial port isn't used other than as a comparison, ntpd has been modified to use the Elan Timer with PPS input to the GPIO. I understand that they do not use pps to serial port, but they compare to ntpd with serial input to show how much better they are. What I object to is that that comparison is cooked. ntpd with pps to serial is a lot better than their graph indicates. They are doing something very wrongly to get such bad results. And it is unnecessary because they are better than ntpd with pps to serial. I have a couple of motherboards with chipsets that have higher resolution timestamping but the GPIO pins aren't accessible. David http://www.synclab.org/radclock/ An alternative to ntpd David ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] 4.2.4p8 - 4.2.6p5 less frequent driftfile updates?
Terje Mathisen writes: Jochen Bern wrote: Hello everyone, about 40h ago, I updated half a platform from CentOS 6.4 to 6.5, which includes updating ntpd from ntp-4.2.4p8-3.el6.centos.x86_64 to ntp-4.2.6p5-1.el6.centos.x86_64. One of the updated machines holds the platform's Internet-bound NTP connections, and has some extra monitoring for that - which now alerts us to the fact that by now, the driftfile gets updated only every 6+ hours, rather than every hour (modulo sync problems) as before. A quick scan seems to confirm that the update frequency is tied to the ntpd version running on the machines: This is intentional, in order to lace less load on the hard drive/flash drive of embedded/small servers. :-) The update is written only if the delta in time or frequency is large enough afair. Yes, this change went in on 10 June 2007, as part of DLM's leapfile improvements, and were in 4.2.5p47. Also see the nonvolatile directive in the ntp.conf file (described in the miscopt.html page). H ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] NTP server with Garmin GPS 18 LVC
unruh wrote: On 2013-12-19, David Lord sn...@lordynet.org wrote: unruh wrote: On 2013-12-19, David Lord sn...@lordynet.org wrote: Adrian P wrote: . Everywhere is written that FreeBSD is the best OS for time keeper, since its kernel supports nanosecond accuracies. However, if GPS 18x LVC PPS signal has only microsecond accuracy, then I suppose there is no point on using FreeBSD, any linux would be okay as well. So what do you think, looking from NTP point of view only, is this nanosecond accuracy the only reason why someone would use FreeBSD, or there are some other reasons as well? I see you have experience with all kind of OSes. I think the combination of hardware used is at least as important as the OS. http://www.febo.com/pages/soekris/ 100ns accuracy AGreed. But their comparison is strongly cooked. At least on my system, the serial port interrupt performace is far better than theirs-- something like 2 us, not the 50us they are showing. Certainly 100ns is better, but more like one order of magnitude, not more than 2. It might not be very clear but PPS to serial port isn't used other than as a comparison, ntpd has been modified to use the Elan Timer with PPS input to the GPIO. I understand that they do not use pps to serial port, but they compare to ntpd with serial input to show how much better they are. What I object to is that that comparison is cooked. ntpd with pps to serial is a lot better than their graph indicates. They are doing something very wrongly to get such bad results. And it is unnecessary because they are better than ntpd with pps to serial. My loop_summary from Sure GPS + NetBSD-6 + ntp-dev-4.2.7p401 over a 7 day period has offset(us) 7 +/-34 to 21 +/- 52 and rms offset from 3.9 to 6.1. If ntpd is throwing away 7 out of 8 polls I'd say their results could be similar to mine. Also since from that page ntp gettime Thu, Oct 19 2006 20:10 there have been many changes to ntpd. David ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] NTP server with Garmin GPS 18 LVC
David Lord writes: If ntpd is throwing away 7 out of 8 polls I'd say their results could be similar to mine. ntpd doesn't throw away 7 out of 8 polls, Bill just likes to say that. ntpd chooses the best data it sees (for its definition of best) and uses those. H ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions