Re: [ntp:questions] Red Hat vote for chrony
On Mon, 8 Dec 2014, Phil W Lee wrote: In theory, this wouldn't be expensive if done at the mass production stage, but clock stability isn't high enough on the design priorities for designers to put it into mass market machines. It's been awhile since I've heard this whine on c.p.t.n. Complaints about cheap crystals in PC-compatibles used to be quite common, but they seemed to have stopped for a while right after my intention to respond to them crystallized (no pun intended). I don't think a computer with a better IRQ0 timer would be that useful. NTP has survived quite well on what we have, and any serious clock nerd uses external PPS anyway. After all, a lot more people will pay extra for a rubidium module in their GPS, than for one in their desktop There are better ways to upgrade a computer design for timekeeping. For one thing, computers these days have dropped the RS-232 and parallel connectors that used to be usable as a geekport to get PPS signals in with lower and more consisent latency than USB. Putting some sort of geekport back in should be a higher priority than a better crystal. The think *I* find most maddening about the PC design is that, ever since the AT, there has been a circuit especially put there to track real time -- the CMOS clock. But it has many flaws, so NTP and the OS ignore it except at boot-up. Change it to count time in 64-bit NTP format and allow subsecond reads, writes and trimming and then it would become quite useful. Especially since it would free the OS to do anything it wants with the IRQ0 timer without disrupting wallclock time. Tickless kernels have been a thorn in the side of clock nerds, but this would avert the conflict. Michael Deutschmann mich...@talosis.ca ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Wed, Dec 10, 2014 at 5:04 AM, Miroslav Lichvar mlich...@redhat.com wrote: See also the following post, apparently the BBB can use a separate HW timer to timestamp PPS, reducing the error and jitter significantly. http://blog.dan.drown.org/beaglebone-black-timer-capture-driver/ It looks like this could be a really nice and cheap NTP/PTP server. Here's Dan's post to time-nuts about this https://www.febo.com/pipermail/time-nuts/2014-December/089217.html. He provided two more graphs. Apropos our subject he uses Chrony. There are some other people doing PRU and timer capture on the Black as well as external clocking. Hence my earlier reference to it if one wanted to continue in the spirit of the 4501 mods. Only without modifying a motherboard. Of course Ackerman's graphs makes it obvious that having nanosecond errors on your servers is irrelevant once you're outside the box. People are also working on PTP so you might actually be able to build a nice clock for $250 in materials. Here's Simon Marsh's hardware mod a la the PHK/Ackerman 4501 on the BBB if you want skip external clocking. https://www.febo.com/pipermail/time-nuts/2014-November/088385.html ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On 09/12/14 07:06, Charles Swiger wrote: Yes, I also find it a bit surprising than modern desktop CPUs and GPUs are willing to run right up to their thermal trip points of ~80 C or so rather than bump up fan speed a little more to keep them more around 50 C. Cost engineering, for what is a throwaway product with a fashion life of less than 3 years. Also, Older systems tended to use more aggressive cooling, especially laptops. Well, smarter firmware and Hall effect sensors to measure fan speed means you can spin the fans more slowly than if you needed to apply 40% minimum speed just to be sure that the fan would spin up from idle. Variable speed fans are available now and motherboards support them. Tachometer outputs have been around for a long time, and because fans use brushless DC motors (i.e. synchronous motors with electronics to generate the AC) they all have to have rotational phase detectors (although it may be done by monitoring the coil current, rather than a separate sensor). Heavy use of these tends to be associated with quiet PCs, so the BIOS may well be set for minimal cooling, rather than keeping a low temperature. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On 2014-12-09, Charles Swiger cswi...@mac.com wrote: ... [ ... ] Of course, in an ideal system designed for clock stability, the clock variation with temperature would be measured, and an offset based on the temperature of the crystal applied directly to the output of the clock itself, before it was supplied to the operating system. Yes; you're describing calibrating a temperature-compensated XO, or TCXO. There are also versions of ntp which have a temp compensation/measurement system compiled in to apply to the clocks. It does tend to give much better control of the clock than regular ntpd apparently. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
William Unruh wrote: On 2014-12-09, Charles Swiger cswi...@mac.com wrote: ... [ ... ] Of course, in an ideal system designed for clock stability, the clock variation with temperature would be measured, and an offset based on the temperature of the crystal applied directly to the output of the clock itself, before it was supplied to the operating system. Yes; you're describing calibrating a temperature-compensated XO, or TCXO. There are also versions of ntp which have a temp compensation/measurement system compiled in to apply to the clocks. It does tend to give much better control of the clock than regular ntpd apparently. It does help: On motherboards with a temperature sensor close to the master crystal, you can get somewhere in the 2-10x range improvement in the size of temperature excursions. The correct solution is of course to not depend on $0.10 crystals as the time base for dedicated NTP servers. :-) Terje -- - 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] Red Hat vote for chrony
On Dec 9, 2014, at 2:41 AM, Terje Mathisen terje.mathi...@tmsw.no wrote: [ ... ] Yes; you're describing calibrating a temperature-compensated XO, or TCXO. There are also versions of ntp which have a temp compensation/measurement system compiled in to apply to the clocks. It does tend to give much better control of the clock than regular ntpd apparently. It does help: On motherboards with a temperature sensor close to the master crystal, you can get somewhere in the 2-10x range improvement in the size of temperature excursions. I'd agree with this, although the best case is probably not quite an order of magnitude, more like a factor of 5x. Or perhaps I shouldn't be too optimistic about how bad a really cheap crystal can be. :-) The correct solution is of course to not depend on $0.10 crystals as the time base for dedicated NTP servers. :-) Well, yes. You can get a PCI(e) card with a TCXO or OCXO and an optional GPS module like the Beagle ClockCard or a SpectraCom TSync for a few hundred bucks. That's quite a bit more than a $40 GPS puck, but these will also freewheel for a lot longer before losing or gaining a second in error: ~2 seconds/month if kept stable at 23C, I believe one said. Regards, -- -Chuck ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On 2014-12-09, Charles Swiger cswi...@mac.com wrote: On Dec 9, 2014, at 2:41 AM, Terje Mathisen terje.mathi...@tmsw.no wrote: [ ... ] Yes; you're describing calibrating a temperature-compensated XO, or TCXO. There are also versions of ntp which have a temp compensation/measurement system compiled in to apply to the clocks. It does tend to give much better control of the clock than regular ntpd apparently. It does help: On motherboards with a temperature sensor close to the master crystal, you can get somewhere in the 2-10x range improvement in the size of temperature excursions. I'd agree with this, although the best case is probably not quite an order of magnitude, more like a factor of 5x. Or perhaps I shouldn't be too optimistic about how bad a really cheap crystal can be. :-) The correct solution is of course to not depend on $0.10 crystals as the time base for dedicated NTP servers. :-) Well, yes. You can get a PCI(e) card with a TCXO or OCXO and an optional GPS module like the Beagle ClockCard or a SpectraCom TSync for a few hundred bucks. That's quite a bit more than a $40 GPS puck, but these will also freewheel for a lot longer before losing or gaining a second in error: ~2 seconds/month if kept stable at 23C, I believe one said. I suspect even the cheap ones can do that if kept stable at 23C. (that is about 1PPM) And if you could put a fast thermal probe onto the crystal, you could probably do as well even in a flutuating environment with an addition to ntpd/chrony to use the temp data to compensate the clock rate. Then it would be really useful to keep a long string of data on the offsets and the temp to get a better set of coeficients for the temperature dependence of the rate. Does anyone know in general what fraction of the variablility of those cheap crystals is due to temp, and how much is due to other sources(crystal defect motion for example, or capacitor aging drift). Regards, ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
William Unruh wrote: On 2014-12-09, Charles Swiger cswi...@mac.com wrote: On Dec 9, 2014, at 2:41 AM, Terje Mathisen terje.mathi...@tmsw.no wrote: [ ... ] Yes; you're describing calibrating a temperature-compensated XO, or TCXO. There are also versions of ntp which have a temp compensation/measurement system compiled in to apply to the clocks. It does tend to give much better control of the clock than regular ntpd apparently. It does help: On motherboards with a temperature sensor close to the master crystal, you can get somewhere in the 2-10x range improvement in the size of temperature excursions. I'd agree with this, although the best case is probably not quite an order of magnitude, more like a factor of 5x. Or perhaps I shouldn't be too optimistic about how bad a really cheap crystal can be. :-) The correct solution is of course to not depend on $0.10 crystals as the time base for dedicated NTP servers. :-) Well, yes. You can get a PCI(e) card with a TCXO or OCXO and an optional GPS module like the Beagle ClockCard or a SpectraCom TSync for a few hundred bucks. That's quite a bit more than a $40 GPS puck, but these will also freewheel for a lot longer before losing or gaining a second in error: ~2 seconds/month if kept stable at 23C, I believe one said. I suspect even the cheap ones can do that if kept stable at 23C. (that is about 1PPM) And if you could put a fast thermal probe onto the crystal, you could probably do as well even in a flutuating environment with an addition to ntpd/chrony to use the temp data to compensate the clock rate. Then it would be really useful to keep a long string of data on the offsets and the temp to get a better set of coeficients for the temperature dependence of the rate. Does anyone know in general what fraction of the variablility of those cheap crystals is due to temp, and how much is due to other sources(crystal defect motion for example, or capacitor aging drift). My xtal book gives either parabolic or lazy-s curves but the real problem with cheap crystals is that the turning point or flat sections can be way off the ambient temperature. David Regards, ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Tue, Dec 9, 2014 at 4:02 PM, David Lord sn...@lordynet.org wrote: My xtal book gives either parabolic or lazy-s curves but the real problem with cheap crystals is that the turning point or flat sections can be way off the ambient temperature. phk suggests that modern high-clock systems have better crystals because multiply to gigaherz. I'm not a time-nut so I have no idea. Besides it's not a real problem to solve (yet). ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Dec 9, 2014, at 11:38 AM, William Unruh un...@invalid.ca wrote: Well, yes. You can get a PCI(e) card with a TCXO or OCXO and an optional GPS module like the Beagle ClockCard or a SpectraCom TSync for a few hundred bucks. That's quite a bit more than a $40 GPS puck, but these will also freewheel for a lot longer before losing or gaining a second in error: ~2 seconds/month if kept stable at 23C, I believe one said. I suspect even the cheap ones can do that if kept stable at 23C. (that is about 1PPM) And if you could put a fast thermal probe onto the crystal, you could probably do as well even in a flutuating environment with an addition to ntpd/chrony to use the temp data to compensate the clock rate. Then it would be really useful to keep a long string of data on the offsets and the temp to get a better set of coeficients for the temperature dependence of the rate. Does anyone know in general what fraction of the variablility of those cheap crystals is due to temp, and how much is due to other sources(crystal defect motion for example, or capacitor aging drift). For a specific example, Beagle documents that in their specs, ie: http://www.beaglesoft.com/clcaspecs.htm The precision of the ClockCard is entirely dependent on the quality of the oscillator circuit. There are three sources of error in the oscillator: (1) calibration error, (2) temperature stability, and (3) aging. Understanding these will allow you to estimate the precision of the ClockCard in your application. Calibration error: The ClockCard oscillator is calibrated at the factory to within 1 PPM part per million) of its specified frequency at room temperature (23° C or 73° F). Temperature Stability: The frequency of oscillation of crystal oscillators is highly dependent on temperature. The oscillator used in the ClockCard has an extremely low temperature dependency of 5 PPM from 0° C to 50° C (32° F to 122° F). Since the oscillator is calibrated to 1 PPM at room temperature (23 °C), it will only exhibit 1 PPM error if its environment is held to this temperature. The worst case condition is if the temperature of the ClockCard is held at one of the extremes, 0 or 50 °C. At these points, there will be an error of 5 PPM. If the temperature variation covers a smaller span, less error will be exhibited. Aging: All crystal oscillators have an aging characteristic. The crystal used in the ClockCard uses the coldweld manufacturing technique, which exhibits the lowest aging characteristic of 1 PPM per year. In practice, this aging rate improves significantly with time, but for practical purposes the value of 1 PPM is adequate. ...and that's for a good crystal used as a TCXO. Cheaper crystals are usually going to be worse. If you meant more in general, Wikipedia has collected decent data here: http://en.wikipedia.org/wiki/Crystal_oscillator ...but if you wanted published reference data, perhaps the NIST docs here: http://tf.nist.gov/general/pdf/906.pdf http://tf.nist.gov/general/pdf/581.pdf Regards, -- -Chuck ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
Charles Swiger cswi...@mac.com wrote: The correct solution is of course to not depend on $0.10 crystals as the time base for dedicated NTP servers. :-) Well, yes. You can get a PCI(e) card with a TCXO or OCXO and an optional GPS module like the Beagle ClockCard or a SpectraCom TSync for a few hundred bucks. That's quite a bit more than a $40 GPS puck, but these will also freewheel for a lot longer before losing or gaining a second in error: ~2 seconds/month if kept stable at 23C, I believe one said. But how much better is such a card at stabilizing the real time clock in the system? Aren't these cards ultimately used to lock the cheap crystal in the system itself? And how fast is the control loop there? I am using ntpd (development release) to lock the clock of a Linux system to ntpd, and of course I can observe the excursions that result from temperature increases and decreases. When monitoring the offset measured by ntpd it is clearly visible that there is an offset proportional to the derivative of the temperature. Apparently ntpd is not quick enough to adapt the frequency estimate, and the offset remains nonzero until the temperature again stabilizes. Shortening the poll interval improves this, but there still is an offset of up to 8us when the temperature changes 4degC/hour. (synced to an external GPSDO via PPS) Does such a card do better, or is it in fact just a GPSDO on a PCI card that still is only used to lock the system time via ntpd? ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Sun, 07 Dec 2014 09:46:35 +, David Woolley wrote: On 07/12/14 09:08, detha wrote: More and more of those servers end up being virtualized. Quicker reaction to virtualization funnies, and faster convergence on VMs that are spun up/down on demand, seem to be one of the main reasons Redhat switched to chrony. Surely the right solution for that is a virtualisation assist API that provide access to the host time, and if it is possible to read residual counts/cycle counters, without virtualisation overhead, to communicate the information needed to use that data for the next few seconds. Obviously that requires work from the developers of the virtual hosts. It requires development on both the vhost and the guest systems. The typical OS reads time from the RTC at boot, and then keeps its own time based on clock interrupts/TSC/... To get all those to sync perfectly is apparently harder than it seems, especially when one takes into account that the guest can migrate between vhosts. In any case, it means running some driver on the guest that updates the guest's idea of current time based on a call to the hypervisor. Which is close to running an ntp client on the guest with a special refclock driver that calls to the hypervisor. (come to think of it, do such a refclock drivers exist?) The 'recommended best practice' seems to have been oscillating between 'sync with vhost' versus 'guest runs own (s)ntp client' in a one or two year cycle. Every once in a while VMWare/KVM/Xen think 'now we have got guest sync right', and advise to use guest sync. Then is turns out there are still some corner cases where it doesn't work, and the advise goes back to 'use an ntp client on the guest'. -d ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Mon, Dec 08, 2014 at 03:27:15AM +, William Unruh wrote: On 2014-12-07, Charles Swiger cswi...@mac.com wrote: Yes, so chrony recommends using maxpoll=4 to the LAN, and not only to local refclocks. No, read the chrony docs. the default is maxpoll 10 minpoll 5. The default minpoll is 6 and maxpoll 10, exactly the same as in ntpd. That the faq as an example uses 2 and 4 is I agree stupid. It is faq. I have no idea who wrote it. I wrote it. What exactly is wrong with poll 4 on a LAN? -- Miroslav Lichvar ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Sat, Dec 06, 2014 at 03:35:10PM -0500, Paul wrote: On Sat, Dec 6, 2014 at 11:12 AM, William Unruh un...@invalid.ca wrote: And in my tests 10 years ago or so, I used a local gps clock to test the ability of chrony and ntpd to discipline a computer clock networked to another server which was disciplined by a gps. Thus the network was the same, and the difference was ntpd vs chrony. chrony was better. Primarily, I think, because chrony responded more quickly to drift rate changes due to temp changes. I looked at your data back in the day. Even then I thought they were old. Of course if the secret sauce is loop constants (I haven't read the Chrony architecture document, maybe because there isn't one) then perhaps the results would still be the same. The main part of the secret sauce is the variable number of points used in the linear regression. When the clock frequency changes quickly, only a small number of points will be used to get a better estimate of the current frequency offset. I.e. chronyd adapts to the Allan intercept without changing the polling interval. This adaptation doesn't always work perfectly, the current code often reduces the number of points unnecessarily, but there are some ideas that will likely be implemented in the future to improve it. Of course, a similar approach could be used with the NTP PLL/FLL loop. If the time constant wasn't fixed to the polling interval and the FLL part of the loop wasn't active only when the update interval is longer than 2048 seconds, the performance could be improved significantly. I was suggesting this years ago. It would be nice if there was at least a tinker option to shift the constant as needed, but a patch for that was rejected. So now the Linux kernel uses a nonstandard PLL shift to get a better performance with ntpd and current typical network jitters. If you like the slow response of ntpd, chronyd can be configured with the minsamples and maxsamples options to use a fixed number of points and to some extent imitate ntpd. In my testing, when the number was set to 40 the overall time and frequency errors were quite close to ntpd (running on Linux). -- Miroslav Lichvar ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Dec 7, 2014, at 7:19 PM, William Unruh un...@invalid.ca wrote: I suspect most people are a bit more likely to use air conditioning to control ambient temperature changes then they are to desolder and swap out their crystals in the hopes of obtaining more precise timekeeping Actually air conditioning is largely irrelevant because it is not ambient air temperature changes that are most important, but temp changes inside the case caused by differing loads on the system. Data point: for a normal desktop machine I have at home, which has a 95W TDP i5 CPU and a 145W 970 GPU, I can see a ~12C temperature change on the motherboard temp sensor between idle and full load on both as the total system draw goes from ~90 W to ~350 W, if I don't use air-conditioning. With air-conditioning on, the temperature change shrinks to about 5C, which reduces the thermal wandering of the XO by a factor of 2. That seems to be a worthwhile improvement, not largely irrelevant. Furthermore, most of the systems I deal with at work are either 1U or blades in datacenter racks with the raised floor forming a plenum to deliver temperature-controlled air to each rack, hot and cold isle design, etc. I can't observe even a 1C change in temp just by running an individual machine or VM at peak load. Even firing off something which causes a load spike for an hour or two across all of the systems or VMs in a particular rack only causes a 2-3C change. In practice, that limits thermal wandering due to load from 10-20 PPM to around ~2 PPM. One of my collegues debvised a script whose sole purpose was to stress the cpus so he could use the air coming out to dry his socks. Ie, cpu load drives temp change which produces time shifts. Yes. That matters the most for freestanding machines which are not kept in air-conditioning. It matters very little for machines in a data center, because the ambient thermals there are controlled fairly precisely. And it is there that chrony tends to be better at keeping track of the rate changes. That's the claim chrony makes, yes. Regards, -- -Chuck ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Dec 7, 2014, at 7:27 PM, William Unruh un...@invalid.ca wrote: On 2014-12-07, Charles Swiger cswi...@mac.com wrote: On Dec 6, 2014, at 8:33 AM, William Unruh un...@invalid.ca wrote: On 2014-12-06, Charles Swiger cswi...@mac.com wrote: [ ... ] Dude, give it a rest. You've just acknowledged that the chrony docs at the URL ending with manual.html directly above recommend using maxpoll=4 to the LAN. No, the FAQ has that as an example. Yes, the default values compiled into chrony of minpoll=5 maxpoll=10 are nearly the same as what ntpd uses, which is polite when talking to the NTP pool or other WAN sources that you haven't made prior arrangements with. Can we skip the pendantry involved between default and recommended, especially when you seem to prefer the recommended faster polling yourself? Especially when the document you quote does not use the word recommended It uses the word example. Um, the first sentence of section 5.3.4 uses the word 'recommended'. I was going to quote it yet again for the third time, but I saw that Miroslav just replied saying that he wrote that section. You can argue this point further with him if you chose. [ ... ] When ntpd has been up for a while using default maxpoll=10, how many past polls are available (per timesource), and what interval of time does that represent? You answered that below, in fact: Note that before it actually runs, it does remember the past 8 offsets AND delays, ...which turns into 8 * 1024 seconds ~= 2.27 hours. If we cannot manage to agree that this data is kept in memory by ntpd, then further discussion of what ntpd might or might not do in discarding 7 out of 8 polls and so forth is completely moot. It is not used to set the clock. NOte that I will agree that ntpd also writes to a file the measurements in the logs. So what? It is not used by ntpd. You've acknowledged that it keeps the data around, which means it remembers that information. The source code to clock_combine() was just posted by David Woolley. (Starting from a false premise proves nothing about the following conclusion.) Since you keep wanting to say I am wrong, why do you not tell us how ntpd works in your understanding? Let me provide three responses: 1) I'd prefer you to say things which are accurate as to fact rather than arguing. 2) If you believe a statement I've made to be mistaken, feel free to point it out. 3) I'd rather test and submit patches than try to educate someone who doesn't want to learn. Thought so. I interpret 1 as I don't know, You'd be wrong in that intepretation if that's what you think I meant. 2 as the fewer statements I make the less others can point out. I don't think I've avoided making statements in this thread. Either point out specific issues, if any, or you can make general handwaving notions instead and pretend that you've identified a specific issue without ever identifying one. For 3) how many patches and to what have you submitted? You can google my name and tvtohz, setitimer, or wakeup latency and get results like: http://lists.freebsd.org/pipermail/freebsd-stable/2009-September/051803.html The discussion was specifically including FreeBSD, DragonFlyBSD, Darwin/MacOS, but should be relevant to the other BSD variants like NetBSD and OpenBSD. Regards, -- -Chuck ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On 2014-12-08, Charles Swiger cswi...@mac.com wrote: On Dec 7, 2014, at 7:19 PM, William Unruh un...@invalid.ca wrote: I suspect most people are a bit more likely to use air conditioning to control ambient temperature changes then they are to desolder and swap out their crystals in the hopes of obtaining more precise timekeeping Actually air conditioning is largely irrelevant because it is not ambient air temperature changes that are most important, but temp changes inside the case caused by differing loads on the system. Data point: for a normal desktop machine I have at home, which has a 95W TDP i5 CPU and a 145W 970 GPU, I can see a ~12C temperature change on the motherboard temp sensor between idle and full load on both as the total system draw goes from ~90 W to ~350 W, if I don't use air-conditioning. With air-conditioning on, the temperature change shrinks to about 5C, which reduces the thermal wandering of the XO by a factor of 2. That seems to be a worthwhile improvement, not largely irrelevant. I find that very stange. Does the temperature in the room really oscillate by 5 degrees if you run your machine? (Ie, you ascribe 5C of the temp change to the room). Furthermore, most of the systems I deal with at work are either 1U or blades in datacenter racks with the raised floor forming a plenum to deliver temperature-controlled air to each rack, hot and cold isle design, etc. I can't observe even a 1C change in temp just by running an individual machine or VM at peak load. Even firing off something which causes a load spike for an hour or two across all of the systems or VMs in a particular rack only causes a 2-3C change. And those blades may well have much better internal cooling. In practice, that limits thermal wandering due to load from 10-20 PPM to around ~2 PPM. One of my collegues debvised a script whose sole purpose was to stress the cpus so he could use the air coming out to dry his socks. Ie, cpu load drives temp change which produces time shifts. Yes. That matters the most for freestanding machines which are not kept in air-conditioning. It matters very little for machines in a data center, because the ambient thermals there are controlled fairly precisely. And it is there that chrony tends to be better at keeping track of the rate changes. That's the claim chrony makes, yes. Actually no, I do not think I have read chronyd making that claim. I have made that claim for chrony to try to explain the much better stats that chrony is measured to have. The stats are not a claim, they are measurements. The explanation is a claim. Chrony IS much faster at accounting for fluctuation in the rate or the offset. That I have also measured. Regards, ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Dec 8, 2014, at 9:19 AM, William Unruh un...@invalid.ca wrote: Data point: for a normal desktop machine I have at home, which has a 95W TDP i5 CPU and a 145W 970 GPU, I can see a ~12C temperature change on the motherboard temp sensor between idle and full load on both as the total system draw goes from ~90 W to ~350 W, if I don't use air-conditioning. With air-conditioning on, the temperature change shrinks to about 5C, which reduces the thermal wandering of the XO by a factor of 2. That seems to be a worthwhile improvement, not largely irrelevant. I find that very stange. Does the temperature in the room really oscillate by 5 degrees if you run your machine? (Ie, you ascribe 5C of the temp change to the room). With AC on, the room stays steady at 22 C or maybe goes up by one to 23 C. With AC off and given several hours of load, the room temperature changes from 22 C to maybe 25 C. My computer is located at the opposite corner of the room from the AC; if it were closer, it would probably show less of a swing Regards, -- -Chuck ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Dec 8, 2014, at 12:04 PM, Phil W Lee p...@lee-family.me.uk wrote: With air-conditioning on, the temperature change shrinks to about 5C, which reduces the thermal wandering of the XO by a factor of 2. That seems to be a worthwhile improvement, not largely irrelevant. Of course, server class machines tend to be designed with far tighter control of temperature, since wide temperature fluctuations are the worst enemy of reliability. Indeed, although some servers are designed for SOHO and home media stuff, and folks want those to run a little quieter than the typical datacenter server, which tends to resemble a wind tunnel. :-) [ ... ] Furthermore, most of the systems I deal with at work are either 1U or blades in datacenter racks with the raised floor forming a plenum to deliver temperature-controlled air to each rack, hot and cold isle design, etc. I can't observe even a 1C change in temp just by running an individual machine or VM at peak load. Even firing off something which causes a load spike for an hour or two across all of the systems or VMs in a particular rack only causes a 2-3C change. In practice, that limits thermal wandering due to load from 10-20 PPM to around ~2 PPM. So an order of magnitude better, even when tested to extremes which are well beyond any reasonable expectation of normal use. I would expect normal use probably stays within 1 PPM on those systems, based on the figures you've given for extreme load variations. Yes, exactly. Without doing anything unusual, a machine racked in a DC is likely to have quite reasonable thermal stability. That said, I'll admit that ~1-2 PPM isn't especially accurate; it'd be an error of at least tens of milliseconds per day if the local clock lost connection to a local or networked timesource. That would translate into weeks or even a month or so before the error accumulated to a full second, which isn't too bad from a human timescale perspective. One of my collegues devised a script whose sole purpose was to stress the cpus so he could use the air coming out to dry his socks. Ie, cpu load drives temp change which produces time shifts. Yes. That matters the most for freestanding machines which are not kept in air-conditioning. It matters very little for machines in a data center, because the ambient thermals there are controlled fairly precisely. And the machines themselves are better designed to run at constant temperature irrespective of load. Freestanding office machines have noise reduction well above temperature stability on the list of design priorities. The attitude with them is just to stop it actually overheating to the point of failure, although it may struggle with even that when a bit of ordinary office dust gets into the works. Of course, a server room is unlikely to have very much of that dust, either. Yes, I also find it a bit surprising than modern desktop CPUs and GPUs are willing to run right up to their thermal trip points of ~80 C or so rather than bump up fan speed a little more to keep them more around 50 C. Older systems tended to use more aggressive cooling, especially laptops. Well, smarter firmware and Hall effect sensors to measure fan speed means you can spin the fans more slowly than if you needed to apply 40% minimum speed just to be sure that the fan would spin up from idle. [ ... ] Of course, in an ideal system designed for clock stability, the clock variation with temperature would be measured, and an offset based on the temperature of the crystal applied directly to the output of the clock itself, before it was supplied to the operating system. Yes; you're describing calibrating a temperature-compensated XO, or TCXO. In theory, this wouldn't be expensive if done at the mass production stage, but clock stability isn't high enough on the design priorities for designers to put it into mass market machines. Nope. Generic AT crystals are dirt-cheap so I doubt many folks would pay even $10 extra for a TCXO variant, much less be willing to pay twice as much for their motherboard to get an SC cut crystal in an OCXO. Regards, -- -Chuck ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Sat, 06 Dec 2014 14:27:47 +, David Woolley wrote: On 06/12/14 12:27, Charles Swiger wrote: Some folks will be satisfied with Miroslav Lichvar's simulations but I'm not smart enough to understand them. Other's will be convinced because they can run Chrony on a Unix laptop and save power or discipline a clock with only a few minutes a day of connect time. I don't know who those people are. Sure, those seem to have been the major use-cases that chrony wanted to handle. RHEL, the subject of this thread, is not intended for laptops; it is intended for mission critical servers. More and more of those servers end up being virtualized. Quicker reaction to virtualization funnies, and faster convergence on VMs that are spun up/down on demand, seem to be one of the main reasons Redhat switched to chrony. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On 07/12/14 09:08, detha wrote: More and more of those servers end up being virtualized. Quicker reaction to virtualization funnies, and faster convergence on VMs that are spun up/down on demand, seem to be one of the main reasons Redhat switched to chrony. Surely the right solution for that is a virtualisation assist API that provide access to the host time, and if it is possible to read residual counts/cycle counters, without virtualisation overhead, to communicate the information needed to use that data for the next few seconds. Obviously that requires work from the developers of the virtual hosts. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Dec 6, 2014, at 11:46 AM, Paul tik-...@bodosom.net wrote: On Sat, Dec 6, 2014 at 7:27 AM, Charles Swiger cswi...@mac.com wrote: Um, certainly? Are you concerned about the quality of the XO which shipped with a Soekris board? That'd be important if you were running a stratum-2+ timeserver. If you don't do the reading you won't understand the problems. I would agree with this point if it were applicable. Here's a salient quote from Ackerman [ ... ] This isn't about modifying a 4501 for testing or measurement. It's about modifying it for high resolution production service. By the way in the current era the 4501 is poor choice. Common SoC based systems like the BeagleBone Black have everything you need (and more), are cheaper and no un-soldering is required. The thread was originally about Red Hat adopting chrony, and how one might measure its timekeeping with sub-microsecond accuracy. You then claimed: But to try and stay on point: how do you measure the performance of a typical disciplined system (virtual) clock in a general purpose computer -- ideally without specialized hardware and software but those are okay if the results can be meaningfully generalized. Given those measures you can decide if you prefer NTP, Chrony, SNTP or something else. ...so the details of using SoC based systems rather than a Net4501 nowadays shouldn't be relevant to what you asked for. Heck, even today's modern smartphone has an AGPS chip and more horsepower than a ~2002-era Soekris with an embedded CPU. You setup a high quality clock somewhere and pull timestamps from the general purpose computer you want to measure. Perhaps you're unfamiliar with the problem PTP solves. That is one of things you have to deal with if you're going to use external measurements to assess the quality of your clock discipline. If you want to use PTP for timestamping, rather than a PPS signal over a GPIO pin, parallel/serial port, fine. What PTP time source did you have in mind? If a $1500 PRS-10 is out-of-bounds even for measurement purposes against a ~$250 embedded system + GPS puck, surely a Symmetricom XLi or Meinberg LANTIME PTP grandmaster clock would also fall outside of your chosen realm of without specialized hardware. Maybe you're also unfamiliar with what NTP is doing: It disciplines a virtual (system) clock which is derived from the on-board (usually uncompensated) crystal. I've been providing NTP stratum-1 or -2 timeservice to the public since the late 80's. I can recall switching to ntpd from timed about the timeframe that physical networks were moving from coax to RJ-45 plugs, anyway. Nevertheless, you're welcome to patronize me about my limited knowledge of NTP if you at least have a productive point to make by doing so. That's why NTP can act as a thermometer. The big problem is temperature induced wander and that's something (back to RedHat and Chrony) Chrony claims to do better. Yes, ntpd and chrony will notice temperature changes causing a frequency drift, at least when a system is using cheap AT-cut XOs, which commonly show a temperature-dependency frequency variation in the 10s of PPM range for a 10 C change around room temperature. A more expensive SC-cut XO will likely remain stable to a few PPM for such a range, and a calibrated TCXO or OCXO will do better still. I suspect most people are a bit more likely to use air conditioning to control ambient temperature changes then they are to desolder and swap out their crystals in the hopes of obtaining more precise timekeeping Regards, -- -Chuck ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Dec 6, 2014, at 8:33 AM, William Unruh un...@invalid.ca wrote: On 2014-12-06, Charles Swiger cswi...@mac.com wrote: On Dec 5, 2014, at 8:39 PM, William Unruh un...@invalid.ca wrote: [ ... ] Just like your claim whether the chrony docs recommend using maxpoll=4 across the network to a LAN timesource or not was wrong? I have given you the quotes from the chrony docs. Where do you get your statement from? The chrony docs, section 5.3.4: http://chrony.tuxfamily.org/manual.html#How-can-I-improve-the-accuracy-of-the-system-clock-with-NTP-sources_003f OK, lets quote the whole section [ ...section snipped, as you found what I linked to and even quoted before... ] Note the caveates : server located on the same lan. In fact more frequent polling IS better especially when one uses the a system with memory-- ie where the system uses the offsets measured for the last 64 times to figure out what the best time now is. Yes, so chrony recommends using maxpoll=4 to the LAN, and not only to local refclocks. But, compare this with the actual documentation for chrony, not this faq which I quoted, and you erased. (Minpoll 5 maxpoll 10) Dude, give it a rest. You've just acknowledged that the chrony docs at the URL ending with manual.html directly above recommend using maxpoll=4 to the LAN. Yes, the default values compiled into chrony of minpoll=5 maxpoll=10 are nearly the same as what ntpd uses, which is polite when talking to the NTP pool or other WAN sources that you haven't made prior arrangements with. Can we skip the pendantry involved between default and recommended, especially when you seem to prefer the recommended faster polling yourself? Just like your claim about whether ntpd cares about figuring out the local clock frequency or whether it only chases the offset was wrong? ntpd does not care about figuring out the local clock frequency. You keep repeating this assertion when it is quite obviously wrong. Above, you've acknowledged that ntp.drift contains the (frequency) offset for the clock. That's the single most important piece of data to have around even if the machine loses network connectivity with other timeservers, or ntpd is restarted, or the machine itself is rebooted. It is ONLY used when the ntpd is started/restarted. If it makes you feel happy, yes ntpd does know about the current drift rate. Try not to confuse making me happy with being able to state the truth, be honestly mistaken about a point but also acknowledge it once corrected, or chosing to be wilfully dishonest. Nothing personal: everyone faces such a choice It knows nothing about the past drift rate. That is what I call memory. And in normal operation, it does not even know about the current drift rate-- it is the operating system that does that. All ntpd does is to change the current drift rate to try to minimize the offset. IF the operating system is incapable of changing the clock rate, ntpd tries to do it for the operating system. Every second it resets the clock to take into account the error in the rate of the clock. In this case, ntpd remembers the current drift rate. But only the current drift rate. Again, it has no memory of the past. When ntpd has been up for a while using default maxpoll=10, how many past polls are available (per timesource), and what interval of time does that represent? You answered that below, in fact: Note that before it actually runs, it does remember the past 8 offsets AND delays, ...which turns into 8 * 1024 seconds ~= 2.27 hours. If we cannot manage to agree that this data is kept in memory by ntpd, then further discussion of what ntpd might or might not do in discarding 7 out of 8 polls and so forth is completely moot. (Starting from a false premise proves nothing about the following conclusion.) Since you keep wanting to say I am wrong, why do you not tell us how ntpd works in your understanding? Let me provide three responses: 1) I'd prefer you to say things which are accurate as to fact rather than arguing. 2) If you believe a statement I've made to be mistaken, feel free to point it out. 3) I'd rather test and submit patches than try to educate someone who doesn't want to learn. ...and skip the rest as being addressed above. Regards, -- -Chuck ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Sun, Dec 7, 2014 at 2:42 PM, Charles Swiger cswi...@mac.com wrote: I would agree with this point if it were applicable. In fact it is as others have tried to point out to you. ... Heck, even today's modern smartphone has an AGPS chip and more horsepower than a ~2002-era Soekris with an embedded CPU. I have no idea what you're talking about but horsepower is not relevant. If you want to use PTP for timestamping, rather than a PPS signal over a GPIO pin, parallel/serial port, fine. What PTP time source did you have in mind? As as been pointed out to you by others you can't measure error using the network unless you remove the network error. That's why you would use PTP but as also noted measuring across the network is the wrong answer. I wasn't suggesting using PTP. I said it tries to solve *one* of the tings you have to deal with. Maybe you're also unfamiliar with what NTP is doing: It disciplines a virtual (system) clock which is derived from the on-board (usually uncompensated) crystal. I've been providing NTP stratum-1 or -2 timeservice to the public since the late 80's. Nevertheless, you're welcome to patronize me about my limited knowledge of NTP if you at least have a productive point to make by doing so. And I've been doing it since you could peer with fuzzballs. Driving to work every day for 30 years doesn't make you an automotive engineer. The big problem is temperature induced wander and that's something (back to RedHat and Chrony) Chrony claims to do better. Yes, ntpd and chrony will notice temperature changes causing a frequency drift, You don't need to repeat approximately what I said back to me. Back to your favorite source: http://phk.freebsd.dk/time/20141026.html but you won't read it so: The reason most crystals don't behave that way, is that people usually don't put a 100W variable heating element right next to them. If you look at the horizontal part, you can see a lot of small wiggles, that is the aircondition cycling on and off. Around 5000 seconds, there's a dip, that's me using the machine to crunsh a couple of gigabytes of data, and the heat from the CPU affects the crystal. The 100W heater is the CPU by the way. And as long I we're providing pointless qualifications: If a $1500 PRS-10 is out-of-bounds even for measurement purposes against a ~$250 embedded system + GPS puck I have a PRS-10 and s GPSD(D)OCXO system. I even have a decent counter and scope. I didn't say it was out of bounds for measurement I said the opposite. Are you one of those opposite guys? If I say X are you going to say I said not-X? This started when you said: Without measuring the local clock against some other clock or oscillator which is known to be accurate to sub-microsecond levels, one doesn't have the data needed to draw conclusions about the actual timekeeping precision. I (sort of) agree with you but nothing you've said addresses that issue. But you don't have to believe me. Just pay attention to any of the people that have tried to point out your errors. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Dec 7, 2014, at 12:47 PM, Paul tik-...@bodosom.net wrote: On Sun, Dec 7, 2014 at 2:42 PM, Charles Swiger cswi...@mac.com wrote: ... Heck, even today's modern smartphone has an AGPS chip and more horsepower than a ~2002-era Soekris with an embedded CPU. I have no idea what you're talking about but horsepower is not relevant. Oddly enough: if I have no idea what someone was talking about, I don't assume that what they said could not possibly be relevant. If you want to use PTP for timestamping, rather than a PPS signal over a GPIO pin, parallel/serial port, fine. What PTP time source did you have in mind? As as been pointed out to you by others you can't measure error using the network unless you remove the network error. That's why you would use PTP but as also noted measuring across the network is the wrong answer. I wasn't suggesting using PTP. I said it tries to solve *one* of the tings you have to deal with. You've told me one thing that you wouldn't use. That doesn't answer the question of what you would use. Maybe you're also unfamiliar with what NTP is doing: It disciplines a virtual (system) clock which is derived from the on-board (usually uncompensated) crystal. I've been providing NTP stratum-1 or -2 timeservice to the public since the late 80's. Nevertheless, you're welcome to patronize me about my limited knowledge of NTP if you at least have a productive point to make by doing so. And I've been doing it since you could peer with fuzzballs. Driving to work every day for 30 years doesn't make you an automotive engineer. Right, and your point was? Pretend you wanted to explain what you meant by this to an neutral third party. The big problem is temperature induced wander and that's something (back to RedHat and Chrony) Chrony claims to do better. Yes, ntpd and chrony will notice temperature changes causing a frequency drift, You don't need to repeat approximately what I said back to me. I was attempting to find some level of agreement about basic facts. Back to your favorite source: http://phk.freebsd.dk/time/20141026.html but you won't read it so: WTF is that supposed to mean? The reason most crystals don’t behave that way, is that people usually don’t put a 100W variable heating element right next to them. If you look at the horizontal part, you can see a lot of small wiggles, that is the aircondition cycling on and off. Around 5000 seconds, there’s a dip, that’s me using the machine to crunsh a couple of gigabytes of data, and the heat from the CPU affects the crystal. The 100W heater is the CPU by the way. I've either read my supposed favorite source already and/or or not read it. 100W CPU under load causes XO drift. And so? And as long I we're providing pointless qualifications: Um, wow. Did you fail to stroke your ego in private today? Dude, unless I want to hire somebody, I generally don't care any more about someone else's qualifications than I care about what they think of mine. If a $1500 PRS-10 is out-of-bounds even for measurement purposes against a ~$250 embedded system + GPS puck I have a PRS-10 and s GPSD(D)OCXO system. I even have a decent counter and scope. I didn't say it was out of bounds for measurement I said the opposite. Are you one of those opposite guys? If I say X are you going to say I said not-X? If we're talking about data, if X is right, then I'm hopefully going to say X is right. [ ... ] Without measuring the local clock against some other clock or oscillator which is known to be accurate to sub-microsecond levels, one doesn't have the data needed to draw conclusions about the actual timekeeping precision. I (sort of) agree with you but nothing you've said addresses that issue. But you don't have to believe me. Just pay attention to any of the people that have tried to point out your errors. And what errors would these be? Regards, -- -Chuck ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On 2014-12-07, Charles Swiger cswi...@mac.com wrote: ... That's why NTP can act as a thermometer. The big problem is temperature induced wander and that's something (back to RedHat and Chrony) Chrony claims to do better. Yes, ntpd and chrony will notice temperature changes causing a frequency drift, at least when a system is using cheap AT-cut XOs, which commonly show a temperature-dependency frequency variation in the 10s of PPM range for a 10 C change around room temperature. A more expensive SC-cut XO will likely remain stable to a few PPM for such a range, and a calibrated TCXO or OCXO will do better still. I suspect most people are a bit more likely to use air conditioning to control ambient temperature changes then they are to desolder and swap out their crystals in the hopes of obtaining more precise timekeeping Actually air conditioning is largely irrelevant because it is not ambient air temperature changes that are most important, but temp changes inside the case caused by differing loads on the system. One of my collegues debvised a script whose sole purpose was to stress the cpus so he could use the air coming out to dry his socks. Ie, cpu load drives temp change which produces time shifts. And it is there that chrony tends to be better at keeping track of the rate changes. Regards, ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On 2014-12-07, Charles Swiger cswi...@mac.com wrote: On Dec 6, 2014, at 8:33 AM, William Unruh un...@invalid.ca wrote: On 2014-12-06, Charles Swiger cswi...@mac.com wrote: On Dec 5, 2014, at 8:39 PM, William Unruh un...@invalid.ca wrote: [ ... ] Just like your claim whether the chrony docs recommend using maxpoll=4 across the network to a LAN timesource or not was wrong? I have given you the quotes from the chrony docs. Where do you get your statement from? The chrony docs, section 5.3.4: http://chrony.tuxfamily.org/manual.html#How-can-I-improve-the-accuracy-of-the-system-clock-with-NTP-sources_003f OK, lets quote the whole section [ ...section snipped, as you found what I linked to and even quoted before... ] Note the caveates : server located on the same lan. In fact more frequent polling IS better especially when one uses the a system with memory-- ie where the system uses the offsets measured for the last 64 times to figure out what the best time now is. Yes, so chrony recommends using maxpoll=4 to the LAN, and not only to local refclocks. No, read the chrony docs. the default is maxpoll 10 minpoll 5. That the faq as an example uses 2 and 4 is I agree stupid. It is faq. I have no idea who wrote it. But, compare this with the actual documentation for chrony, not this faq which I quoted, and you erased. (Minpoll 5 maxpoll 10) Dude, give it a rest. You've just acknowledged that the chrony docs at the URL ending with manual.html directly above recommend using maxpoll=4 to the LAN. No, the FAQ has that as an example. Yes, the default values compiled into chrony of minpoll=5 maxpoll=10 are nearly the same as what ntpd uses, which is polite when talking to the NTP pool or other WAN sources that you haven't made prior arrangements with. Can we skip the pendantry involved between default and recommended, especially when you seem to prefer the recommended faster polling yourself? Especially when the document you quote does not use the word recommended It uses the word example. Just like your claim about whether ntpd cares about figuring out the local clock frequency or whether it only chases the offset was wrong? ntpd does not care about figuring out the local clock frequency. You keep repeating this assertion when it is quite obviously wrong. Above, you've acknowledged that ntp.drift contains the (frequency) offset for the clock. That's the single most important piece of data to have around even if the machine loses network connectivity with other timeservers, or ntpd is restarted, or the machine itself is rebooted. It is ONLY used when the ntpd is started/restarted. If it makes you feel happy, yes ntpd does know about the current drift rate. Try not to confuse making me happy with being able to state the truth, be honestly mistaken about a point but also acknowledge it once corrected, or chosing to be wilfully dishonest. Nothing personal: everyone faces such a choice It knows nothing about the past drift rate. That is what I call memory. And in normal operation, it does not even know about the current drift rate-- it is the operating system that does that. All ntpd does is to change the current drift rate to try to minimize the offset. IF the operating system is incapable of changing the clock rate, ntpd tries to do it for the operating system. Every second it resets the clock to take into account the error in the rate of the clock. In this case, ntpd remembers the current drift rate. But only the current drift rate. Again, it has no memory of the past. When ntpd has been up for a while using default maxpoll=10, how many past polls are available (per timesource), and what interval of time does that represent? You answered that below, in fact: Note that before it actually runs, it does remember the past 8 offsets AND delays, ...which turns into 8 * 1024 seconds ~= 2.27 hours. If we cannot manage to agree that this data is kept in memory by ntpd, then further discussion of what ntpd might or might not do in discarding 7 out of 8 polls and so forth is completely moot. It is not used to set the clock. NOte that I will agree that ntpd also writes to a file the measurements in the logs. So what? It is not used by ntpd. (Starting from a false premise proves nothing about the following conclusion.) Since you keep wanting to say I am wrong, why do you not tell us how ntpd works in your understanding? Let me provide three responses: 1) I'd prefer you to say things which are accurate as to fact rather than arguing. 2) If you believe a statement I've made to be mistaken, feel free to point it out. 3) I'd rather test and submit patches than try to educate someone who doesn't want to learn. Thought so. I interpret 1 as I don't know, 2 as the fewer statements I make the less others can point out. For 3) how many patches and to what have you submitted? ...and skip the rest as
Re: [ntp:questions] Red Hat vote for chrony
On Dec 5, 2014, at 8:39 PM, William Unruh un...@invalid.ca wrote: This is obviously false. What do you think /etc/ntp.drift is? It is the offset from the standard rate of the clock. That memory is never used except on bootup. ntpd has to know how much to alter the drift. Ah, so you acknowledge that your original statement was wrong. That wasn't intended to be a trick question. Hmm. Unless one considers intellectual honesty to be tricky, perhaps Just like your claim whether the chrony docs recommend using maxpoll=4 across the network to a LAN timesource or not was wrong? I have given you the quotes from the chrony docs. Where do you get your statement from? The chrony docs, section 5.3.4: http://chrony.tuxfamily.org/manual.html#How-can-I-improve-the-accuracy-of-the-system-clock-with-NTP-sources_003f Just like your claim about whether ntpd cares about figuring out the local clock frequency or whether it only chases the offset was wrong? ntpd does not care about figuring out the local clock frequency. You keep repeating this assertion when it is quite obviously wrong. Above, you've acknowledged that ntp.drift contains the (frequency) offset for the clock. That's the single most important piece of data to have around even if the machine loses network connectivity with other timeservers, or ntpd is restarted, or the machine itself is rebooted. Figuring out the local clock frequency is the most important thing ntpd does. Getting that right means that the local clock can be trusted to provide reasonably accurate time for a decent while even if freewheeling. (Or for the folks who don't care whether their noon is a minute off from real time, just so long as everything on the LAN in their isolated island is mutually synced and that there are 3600 computer seconds per hour of wall clock time.) Regards, -- -Chuck ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On 06/12/14 12:27, Charles Swiger wrote: But to try and stay on point: how do you measure the performance of a typical disciplined system (virtual) clock in a general purpose computer -- ideally without specialized hardware and software but those are okay if the results can be meaningfully generalized. Given those measures you can decide if you prefer NTP, Chrony, SNTP or something else. You setup a high quality clock somewhere and pull timestamps from the general purpose computer you want to measure. Something like Nagios This only works for measuring the ability to cope with time source that is an order or magnitude or so worse than your high quality source. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On 06/12/14 12:27, Charles Swiger wrote: Some folks will be satisfied with Miroslav Lichvar's simulations but I'm not smart enough to understand them. Other's will be convinced because they can run Chrony on a Unix laptop and save power or discipline a clock with only a few minutes a day of connect time. I don't know who those people are. Sure, those seem to have been the major use-cases that chrony wanted to handle. RHEL, the subject of this thread, is not intended for laptops; it is intended for mission critical servers. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
Charles Swiger wrote: On Dec 5, 2014, at 5:55 PM, Paul tik-...@bodosom.net wrote: [ ... ] Even back in 2002 with very inexpensive commodity hardware, FreeBSD was able to achieve accuracy measured to ~260 nanoseconds: H. So phk uses a $1,500 rubidium standard as a system oscillator and you call it inexpensive and commodity. No, he used a $1500 rubidium clock to accurately measure the timekeeping quality of a $220 Soekris computer, and concluded: I have earlier complained that no good and cheap hardware were available which could timestamp a PPS signal reliably and precisely but now the Soekris computer has proven that it does indeed deliver just that: With a pricetag of approx USD220 (single unit including enclosure) this is the best hardware you can find for a stratum 1 NTP server. If you wanted to drive such hardware via a ~$40 GPS puck, you'd probably see an accuracy of around a microsecond, perhaps a bit worse depending on the timekeeping accuracy which the GPS puck provides. That was also the level of accuracy I was seeing from generic Intel hardware running FreeBSD as a stratum 1 with a GPS source. I've used a digital frequency counter which had an onboard TCXO (or possibly a DTCXO) for measuring. Although the frequency counter supported receiving higher-quality PPS timing from an external atomic clock, I've never had a Cs or Rb source, so I won't claim to have measured sub-microsecond accuracy with it. He also ran a particular install of BSD and a non-standard NTP. I believe he ran FreeBSD 4.x and likely the ntpd from ports. Regards, Not quite I believe this clock project needed a little work with soldering iron and also ntpd/freebsd had to be modified to support the sc520 gpio. Soekris net4501 with AMD Elan sc520 cpu which has internal time register resolution of around 100n. System clock I believe used a TAPR clock-block frequency synthesizer and a FATpps signal conditioner kits. I couldn't find any UK supplier for the parts and shipping costs from the states was too expensive. David ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On 2014-12-06, Charles Swiger cswi...@mac.com wrote: On Dec 5, 2014, at 8:39 PM, William Unruh un...@invalid.ca wrote: This is obviously false. What do you think /etc/ntp.drift is? It is the offset from the standard rate of the clock. That memory is never used except on bootup. ntpd has to know how much to alter the drift. Ah, so you acknowledge that your original statement was wrong. That wasn't intended to be a trick question. Hmm. Unless one considers intellectual honesty to be tricky, perhaps Just like your claim whether the chrony docs recommend using maxpoll=4 across the network to a LAN timesource or not was wrong? I have given you the quotes from the chrony docs. Where do you get your statement from? The chrony docs, section 5.3.4: http://chrony.tuxfamily.org/manual.html#How-can-I-improve-the-accuracy-of-the-system-clock-with-NTP-sources_003f OK, lets quote the whole section The optimal polling interval depends on many factors, this includes the ratio between the wander of the clock and the network jitter (sometimes expressed in NTP documents as the Allan intercept), the temperature sensitivity of the crystal oscillator and the maximum rate of change of the temperature. An example of the directive for a server located in the same LAN could be server ntp.local minpoll 2 maxpoll 4 polltarget 30 Note the caveates : server located on the same lan. In fact more frequent polling IS better especially when one uses the a system with memory-- ie where the system uses the offsets measured for the last 64 times to figure out what the best time now is. But, compare this with the actual documentation for chrony, not this faq which I quoted, and you erased. (Minpoll 5 maxpoll 10) Just like your claim about whether ntpd cares about figuring out the local clock frequency or whether it only chases the offset was wrong? ntpd does not care about figuring out the local clock frequency. You keep repeating this assertion when it is quite obviously wrong. Above, you've acknowledged that ntp.drift contains the (frequency) offset for the clock. That's the single most important piece of data to have around even if the machine loses network connectivity with other timeservers, or ntpd is restarted, or the machine itself is rebooted. It is ONLY used when the ntpd is started/restarted. If it makes you feel happy, yes ntpd does know about the current drift rate. It knows nothing about the past drift rate. That is what I call memory. And in normal operation, it does not even know about the current drift rate-- it is the operating system that does that. All ntpd does is to change the current drift rate to try to minimize the offset. IF the operating system is incapable of changing the clock rate, ntpd tries to do it for the operating system. Every second it resets the clock to take into account the error in the rate of the clock. In this case, ntpd remembers the current drift rate. But only the current drift rate. Again, it has no memory of the past. Since you keep wanting to say I am wrong, why do you not tell us how ntpd works in your understanding? Figuring out the local clock frequency is the most important thing ntpd does. Getting that right means that the local clock can be trusted to provide reasonably accurate time for a decent while even if freewheeling. But it does not figure out the clock frequency. It just changes the clock frequency to try to eliminate the offset just measured. Note that before it actually runs, it does remember the past 8 offsets AND delays, These are not used in the ntp clock control algorithm. These are used to select the one with the lowest delay, to feed to the algorithm as the current offset, to use in changing the frequency, and the rest are unused. Note that ntpd also keeps itself in memory, so it remembers its own code as well. So to be perfectly clear, ntp does not remember past events to use, in addition to the current event, in adjusting the rate of the clock. That is what I call memory. That is what I call a non-Markovian process. (Or for the folks who don't care whether their noon is a minute off from real time, just so long as everything on the LAN in their isolated island is mutually synced and that there are 3600 computer seconds per hour of wall clock time.) Which chrony and ntpd do well. Regards, ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On 06/12/14 16:33, William Unruh wrote: memory. And in normal operation, it does not even know about the current drift rate-- it is the operating system that does that. All ntpd does is That's because part of the NTP algorithm is implemented in the kernel where it can be done per tick with negligble overhead. to change the current drift rate to try to minimize the offset. It, or the NTP code in the kernel, maintains two separate corrections, one for short term phase errors and one for long term frequency errors. IF the operating system is incapable of changing the clock rate, ntpd tries to do it for the operating system. Every second it resets the clock to take into account the error in the rate of the clock. In this case, ntpd remembers the current drift rate. But only the current drift rate. Again, it has no memory of the past. The very long time constant means that it remembers many hours into the past when operating at the longest poll interval. In fact your real complaint is not that it doesn't use historical data, but that it looks too far back in history. If I read https://www.eecis.udel.edu/~mills/ntp/html/discipline.html correctly, and allowing for times really meaning plus, the frequency control time constant is 2048 seconds (34 minutes) for poll=6 and about 9 hours for poll 10. As noted in that paper, there are two correction terms, a relatively fast one to correct the phase, and the one discussed above to correct the long term frequency. I suspect that chrony does work well in a workstation environment, where temperature is not well controlled and loading is bursty, but will lose some of its advantages in the server environment, at which RHEL is aimed. (Actually, a thought struck me that the real problem with ntpd may be that the time constant for the phase correction (and poll rate) is tied to that for the frequency correction.) ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Sat, Dec 6, 2014 at 7:27 AM, Charles Swiger cswi...@mac.com wrote: Um, certainly? Are you concerned about the quality of the XO which shipped with a Soekris board? That'd be important if you were running a stratum-2+ timeserver. If you don't do the reading you won't understand the problems. Here's a salient quote from Ackerman *To implement high-performance timekeeping, you need to modify the net4501 board to accept an external clock signal*, as well as use the high resolution timer. These steps require fairly good soldering skills as the surface mount components on the board are tiny; if you're unsure of yourself, find someone with SMT soldering experience to help you. Locate and remove the Soekris clock crystal (X1). A small heat gun is the best way to do this, but with care a soldering iron and solder wick can be used to lift the part. The only pin on X1 that we'll use is the one in the upper left corner, so make sure you don't damage that solder pad. And regarding the Clock-Block The Clock-Block is a flexible frequency synthesizer that can be used for many timing and clocking purposes.It accepts an *external reference signal* from 2 MHz to 50 MHz, and can be programmed via DIP switches to generate output frequencies in the range of about 500 kHz to 250 MHz. The output is a square wave at either 3.3 or 5 volts peak-to-peak. An on-board divider circuit allows the output frequency to be reduced by factors of 16 to 16384 for specialized applications.* One application for the Clock-Block is to replace the crystal oscillator of a PC motherboard to allow synchronizing the PC's clock to an external high-precision reference*. This isn't about modifying a 4501 for testing or measurement. It's about modifying it for high resolution production service. By the way in the current era the 4501 is poor choice. Common SoC based systems like the BeagleBone Black have everything you need (and more), are cheaper and no un-soldering is required. You setup a high quality clock somewhere and pull timestamps from the general purpose computer you want to measure. Perhaps you're unfamiliar with the problem PTP solves. That is one of things you have to deal with if you're going to use external measurements to assess the quality of your clock discipline. Maybe you're also unfamiliar with what NTP is doing: It disciplines a virtual (system) clock which is derived from the on-board (usually uncompensated) crystal . That's why NTP can act as a thermometer. The big problem is temperature induced wander and that's something (back to RedHat and Chrony) Chrony claims to do better. I suspect fiddling something into an external TIC is a way to go but something both simpler and more clever is what I'm curious about. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Sat, Dec 6, 2014 at 11:12 AM, William Unruh un...@invalid.ca wrote: You do not understand them, but feel confident to make recommendations despite that? While I'm not making a broader statement about who understands what I believe you're conflating my comments with those from cswi...@mac.com. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Sat, Dec 6, 2014 at 11:12 AM, William Unruh un...@invalid.ca wrote: And in my tests 10 years ago or so, I used a local gps clock to test the ability of chrony and ntpd to discipline a computer clock networked to another server which was disciplined by a gps. Thus the network was the same, and the difference was ntpd vs chrony. chrony was better. Primarily, I think, because chrony responded more quickly to drift rate changes due to temp changes. I looked at your data back in the day. Even then I thought they were old. Of course if the secret sauce is loop constants (I haven't read the Chrony architecture document, maybe because there isn't one) then perhaps the results would still be the same. Other's will be convinced because they can run Chrony on a Unix laptop and save power or discipline a clock with only a few minutes a day of connect time. I don't know who those people are. Both chrony and ntpd send out the same packets and both have roughly the same polls, so where is this powersaving coming from?) And I thought you were chugging the kool-aid: Using chrony over ntp has also other advantages: - smaller memory footprint (1.3MB vs 6MB resident size) - no unnecessary process wakeups, this is good for powersaving. The ntpd process normally wakes up every second. ... For me it's consistency. I want to run the same timekeeping software everywhere* and Chrony doesn't run everywhere. *my everywhere is obviously not anyone else's everywhere. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On 2014-12-06, Paul tik-...@bodosom.net wrote: On Sat, Dec 6, 2014 at 11:12 AM, William Unruh un...@invalid.ca wrote: And in my tests 10 years ago or so, I used a local gps clock to test the ability of chrony and ntpd to discipline a computer clock networked to another server which was disciplined by a gps. Thus the network was the same, and the difference was ntpd vs chrony. chrony was better. Primarily, I think, because chrony responded more quickly to drift rate changes due to temp changes. I looked at your data back in the day. Even then I thought they were old. Of course if the secret sauce is loop constants (I haven't read the Chrony architecture document, maybe because there isn't one) then perhaps the results would still be the same. They are old, and naive. A much better job could be done. However, they are indicative. As I mentioned it would be a really good project to have someone do a better comparison-- both simulated for example with Lichvar's simulation software, and real life measurements. ... Using chrony over ntp has also other advantages: - smaller memory footprint (1.3MB vs 6MB resident size) - no unnecessary process wakeups, this is good for powersaving. The ntpd process normally wakes up every second. That is to impliment the in-house frequency scaling on systems where userland has no way of changing the frequency of the system clock. Under linux and adjtimex one can alter the system clock (by up to 10% of nominal,-- ntpd only uses max 500PPM, while chrony can use the full range) I do not know if ntpd shuts off that once per sec if the system clock can be altered in the kernel. ... For me it's consistency. I want to run the same timekeeping software everywhere* and Chrony doesn't run everywhere. True and fair enough. Chrony only works on Linux or bsd systems. (That could probably be expanded to include OsX but I do not know if it does out of the box). It does not do Windows, so if you have windows machines, chrony is not an option. But then this thread was about Redhat, which is Linux. *my everywhere is obviously not anyone else's everywhere. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Sat, Dec 6, 2014 at 4:57 PM, William Unruh un...@invalid.ca wrote: But then this thread was about Redhat, which is Linux. I understand that but it's not the point. This thread is about a distro defaulting to Chrony -- and why that's a bad (or good) thing. I say it's an inconvenience because I'm uninterested in Chrony on Linux (RedHat or otherwise). While we have many Linux boxes many of our servers are not Linux/BSD based. It's bad enough dealing with various versions of NTP without having yet another critical part of the infrastructure be different. Luckily we're busy moving away from RedHat. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On 2014-12-06, Paul tik-...@bodosom.net wrote: On Sat, Dec 6, 2014 at 4:57 PM, William Unruh un...@invalid.ca wrote: But then this thread was about Redhat, which is Linux. I understand that but it's not the point. This thread is about a distro defaulting to Chrony -- and why that's a bad (or good) thing. I say it's an inconvenience because I'm uninterested in Chrony on Linux (RedHat or otherwise). While we have many Linux boxes many of our servers are not Linux/BSD based. It's bad enough dealing with various versions of NTP without having yet another critical part of the infrastructure be different. Luckily we're busy moving away from RedHat. AFAIK they will still ship ntpd as well. You can pretty easily uninstall chrony and install ntpd if that is what you desire. You have to set up lots of other stuff on your systems anyway. And this default is way way down on the list of problems in comparison to Redhat defaulting to systemd:-) ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On 2014-12-05, Charles Swiger cswi...@mac.com wrote: On Dec 4, 2014, at 7:00 PM, William Unruh un...@invalid.ca wrote: [ ... ] Actually Miroslav Lichvar IS an expert. He is the chrony maintainer, has done a lot of testing comparing chrony to ntpd ( which showed that chrony controlled the clock a factor of 2 to 20 times better than ntpd did), and is with Redhat. The data I've seen for chrony suggests it handles broken clocks such as commonly found in VMs better than ntpd does. The tradeoff is that chrony prioritizes chasing the reference time over first trying to ensure that the local clock frequency is stable, whereas ntpd really wants to make sure that the local clock counts 3600 seconds in each hour of wall-clock time and then worries about slewing the local time to match up with the reference time. Nope. ntp changes the rate of the local clock to correct offsets. That is all it does. It does not make the rate correct, and then the offset. It simply alters the rate at any time so as to decrease the offset, and it does this measurement by measurement. It has no memory. Chrony uses the last N measurements to make the best estimate of the rate and the offset that it can. It sets the rate to the best estimate of the true rate, and then adds a rate correction to decrease the offset. It's informative to note that the chrony docs (section 5.3.4) recommend using minpoll=2 and maxpoll=4! With those settings chrony will send 225 that is for refclocks. polls per hour, versus 3.5 polls per hour for ntpd with its maxpoll=10. Assuming arguendo the claim of a factor of 20 times better is true, I still don't care to pay the price of a factor of 64 times more network polls. You may set the minpoll and maxpoll to whateever you want. But chrony does not advocate a maxpoll of 4 over the network. Read again. Furthermore-- unfortunately-- I have yet to see data on the accuracy of chrony measured against high-quality TCXO or Rb/Cs reference clocks, such as the PRS-10 that PHK used: http://www.thinksrs.com/products/PRS10.htm ...the current version of which claims to have a +/- 10 ns accuracy for the PPS signal. Instead, most of the data I've seen provided for chrony has involved comparing local clock timestamps to the reference timesource or to some other network timesource, without detailed information as to the accuracy of those references. Nope. I have done measurements on the net where I compared the net to a gps PPS source. The computer PPS has an accuracy of about 1microsec and that can be compared to the network time. I get an increase of about 2-3 times better than ntp. Lichvar got something like 20 times better using a PPS against a local high accuracy time source. The main reason seems to be that chrony is far more algile-- it catches temperature drifts much more quickly than ntpd does, for the same poll. Remember that ntpd throws out 85% of the measurements it makes, in order to try (poorly) to compensate for network up-down inequalities. Sometimes, if the network is very variably assymetric that can improve results. Usually it simply throws away valuable measurements. ntpd is also designed to act very slowly to changes in rate. It is a design philosophy Mills defends strongly, with the matra of stability. Chrony, because of its long term memory, recognizes and responds to rate variations much more quickly, with no sign of instability. It would be good to have a detailed analysis of the chrony algorithm to see if there are corner cases where chrony does worse by going unstable. ntpd is simple to analyse (if one ignores the extreme non-linearity of the jump if the offset is greater than 128ms.) If you would like to compare chrony vs ntp with your PRS10, please do so. Otherwise look at some of he numbers I have on www.theory.physics.ubc.ca/chrony Of course you're not going to see much delta between the local clock and the reference that you're polling every 16 seconds. Without measuring the local clock against some other clock or oscillator which is known to be accurate to sub-microsecond levels, one doesn't have the data needed to draw conclusions about the actual timekeeping precision. Actually not true. How do you think the standards of the various contries determine the accuracy of their clocks. They have no better time standard to compare them with. And yet they confidently will quote accuracy figures for their clocks. Study that. Regards, ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Dec 5, 2014, at 3:42 AM, William Unruh un...@invalid.ca wrote: On 2014-12-05, Charles Swiger cswi...@mac.com wrote: On Dec 4, 2014, at 7:00 PM, William Unruh un...@invalid.ca wrote: [ ... ] Actually Miroslav Lichvar IS an expert. He is the chrony maintainer, has done a lot of testing comparing chrony to ntpd ( which showed that chrony controlled the clock a factor of 2 to 20 times better than ntpd did), and is with Redhat. The data I've seen for chrony suggests it handles broken clocks such as commonly found in VMs better than ntpd does. The tradeoff is that chrony prioritizes chasing the reference time over first trying to ensure that the local clock frequency is stable, whereas ntpd really wants to make sure that the local clock counts 3600 seconds in each hour of wall-clock time and then worries about slewing the local time to match up with the reference time. Nope. ntp changes the rate of the local clock to correct offsets. That is all it does. It does not make the rate correct, and then the offset. If you don't know what the rate of the local clock is, how can you figure out the proper slewing rate? One can obviously keep slewing the clock towards the reference time even without having a good idea of the intrinsic drift, but such an approach will tend to overshoot the ideal frequency adjustment and ring or oscillate back and forth. It simply alters the rate at any time so as to decrease the offset, and it does this measurement by measurement. It has no memory. This is obviously false. What do you think /etc/ntp.drift is? Furthermore, I distinctly recall you complaining that ntpd's clock filter throws away 7 out of 8 polls; even though you are mis-interpreting the situation, you at least acknowledged that ntpd is keeping track of prior data. Or did someone else write Message-id: Hh0tu.8666$tr7.4...@fx22.iad? Chrony uses the last N measurements to make the best estimate of the rate and the offset that it can. It sets the rate to the best estimate of the true rate, and then adds a rate correction to decrease the offset. Yes, that's a reasonable approach. It's informative to note that the chrony docs (section 5.3.4) recommend using minpoll=2 and maxpoll=4! With those settings chrony will send 225 that is for refclocks. polls per hour, versus 3.5 polls per hour for ntpd with its maxpoll=10. Assuming arguendo the claim of a factor of 20 times better is true, I still don't care to pay the price of a factor of 64 times more network polls. You may set the minpoll and maxpoll to whateever you want. But chrony does not advocate a maxpoll of 4 over the network. Read again. I suggest following your own advice before trying to correct others. http://chrony.tuxfamily.org/manual.html#How-can-I-improve-the-accuracy-of-the-system-clock-with-NTP-sources_003f 5.3.4 How can I improve the accuracy of the system clock with NTP sources? Select NTP servers that are well synchronised, stable and close to your network. It’s better to use more than one server, three or four is usually recommended as the minimum, so chronyd can detect falsetickers and combine measurements from multiple sources. [ ... ] The optimal polling interval depends on many factors, this includes the ratio between the wander of the clock and the network jitter (sometimes expressed in NTP documents as the Allan intercept), the temperature sensitivity of the crystal oscillator and the maximum rate of change of the temperature. An example of the directive for a server located in the same LAN could be server ntp.local minpoll 2 maxpoll 4 polltarget 30 The docs aren't talking about reference clocks here, they are talking about polling another machine over the LAN. Furthermore-- unfortunately-- I have yet to see data on the accuracy of chrony measured against high-quality TCXO or Rb/Cs reference clocks, such as the PRS-10 that PHK used: http://www.thinksrs.com/products/PRS10.htm ...the current version of which claims to have a +/- 10 ns accuracy for the PPS signal. Instead, most of the data I've seen provided for chrony has involved comparing local clock timestamps to the reference timesource or to some other network timesource, without detailed information as to the accuracy of those references. Nope. I have done measurements on the net where I compared the net to a gps PPS source. The computer PPS has an accuracy of about 1microsec and that can be compared to the network time. About a microsecond is two orders of magnitude worse than ~10 nanoseconds. As I said before, I'd be happy to review data for chrony taken against that quality of reference. I get an increase of about 2-3 times better than ntp. Lichvar got something like 20 times better using a PPS against a local high accuracy time source. The main reason seems to be that chrony is far more algile-- it catches temperature drifts much more quickly than ntpd does, for the same poll. More agile is
Re: [ntp:questions] Red Hat vote for chrony
On 05/12/14 08:42, William Unruh wrote: Nope. ntp changes the rate of the local clock to correct offsets. That is all it does. It does not make the rate correct, and then the offset. Not true. It applies a correction for the offset, such that if that were the only error, it would be eliminated exponentially with a time constant that is related to the poll rate. I believe it expects to update this correction well before it is fully applied. At the same time, it applies a correction to the long term estimate of the frequency, with a much longer time constant. much more quickly than ntpd does, for the same poll. Remember that ntpd throws out 85% of the measurements it makes, in order to try (poorly) to If the round trip variability was sufficiently low that this pre-filter were not needed, any variations in the samples it is ignoring would be largely suppressed by the loop filters, which have time constants that are significantly longer than even eight times the poll interval. ntpd is heavily oversampling, which means that the min and maxpols quoted for chrony imply an even higher excess rate. Actually not true. How do you think the standards of the various contries determine the accuracy of their clocks. They have no better time standard to compare them with. And yet they confidently will quote accuracy figures for their clocks. Study that. I presume that whatever time transfer mechanism they use has well characterised errors. NTP operates in an environment where the errors are not well characterised. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Fri, Dec 5, 2014 at 9:37 AM, Charles Swiger cswi...@mac.com wrote: I also make sure that my timeservers are running in temperature-controlled environments so that such daily drifts you mention are minimized. I'm starting to think that people answering questions are unsure of the real question so they make a number of assumptions. If you care about sub-millisecond time then you need to say that and the question should be answered in that context. I suspect most of the questions here refer to sub-second accuracy and most of the elaboration is unneeded. If all your external clocks fail I suspect the typical user can depend on the disciplined virtual clock for days. For almost all of human history, the sun or the fixed celestial heavens have provided the most accurate time reference available. Even today, we add (or subtract, in theory) leap seconds in order to keep UTC and UT1 aligned to better than a second courtesy of IERS. Yes, the USNO, CERN, and so forth now do have sufficiently high quality atomic clocks which have better timekeeping precision than celestial observations. I think there's some confusion here. Search for BIPM paper clock or read http://www.ggos-portal.org/lang_en/GGOS-Portal/EN/Topics/Services/BIPM/BIPM.html Such a point is orthogonal to the notion of how to measure a local clock I think this is an interesting question. How does one get high resolution measurements of the error in the virtual clock maintained with NTP (or Chrony)? I thought it was done with purpose built systems. I don't expect a random version of Linux on generic hardware to be able to maintain the clock at nanosecond scale. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On 2014-12-05, Charles Swiger cswi...@mac.com wrote: On Dec 5, 2014, at 3:42 AM, William Unruh un...@invalid.ca wrote: On 2014-12-05, Charles Swiger cswi...@mac.com wrote: On Dec 4, 2014, at 7:00 PM, William Unruh un...@invalid.ca wrote: [ ... ] Actually Miroslav Lichvar IS an expert. He is the chrony maintainer, has done a lot of testing comparing chrony to ntpd ( which showed that chrony controlled the clock a factor of 2 to 20 times better than ntpd did), and is with Redhat. The data I've seen for chrony suggests it handles broken clocks such as commonly found in VMs better than ntpd does. The tradeoff is that chrony prioritizes chasing the reference time over first trying to ensure that the local clock frequency is stable, whereas ntpd really wants to make sure that the local clock counts 3600 seconds in each hour of wall-clock time and then worries about slewing the local time to match up with the reference time. Nope. ntp changes the rate of the local clock to correct offsets. That is all it does. It does not make the rate correct, and then the offset. If you don't know what the rate of the local clock is, how can you figure out the proper slewing rate? ntpd does not care what the rate is. If it sees an offset, it alters the rate todecrease that offset. It keeps doing this until the offset is gone. It is designed so that the overshoot is very small (critical damping). One can obviously keep slewing the clock towards the reference time even without having a good idea of the intrinsic drift, but such an approach will tend to overshoot the ideal frequency adjustment and ring or oscillate back and forth. Not if it is properly designed (critical damping). Mills did a good job of designing a feedback system. As he says it is standard engineering practice. It simply alters the rate at any time so as to decrease the offset, and it does this measurement by measurement. It has no memory. This is obviously false. What do you think /etc/ntp.drift is? It is the offset from the standard rate of the clock. That memory is never used except on bootup. ntpd has to know how much to alter the drift. Furthermore, I distinctly recall you complaining that ntpd's clock filter throws away 7 out of 8 polls; even though you are mis-interpreting the situation, you at least acknowledged that ntpd is keeping track of prior data. But it does not use them. It looks at the delays, and choses the shortest delay from the last 8 offset measurements. The scheme is a simple feedback system, which remembers the current drift and measures the current offset. That by current it means the offset with the lowest delay of the last 8 measurements does not alter the design. It is a Markovian system. It does not remember what the drift or the offset was 15 measurements ago, for example. If you really want to understand ntpd, read the documentation or Mill's book. Don't try picking apart a clearly very short descritption. delta r_i= alpha (remotetime_i - localtime_i) is the equation used, where delta r_i is the change in the rate of the clock, remotetime_i is the time as measured by the ntp exchange procedure of he remote clock, and localtime_i is the time on the local clock when that remote time was measured. alpha is a constant chosed to make the feedback loop be close to critically damped. In order to make the system resistant to varying assymetric travel times, they introduced another step. The i in the above is not every measurement, but the best of 8 (ie the one with the shortest delay, under the assumption that that will be the one with least assymetry in the travel path). But it also means that 85 % of the measurements are simply discarded-- never used to determine the clock offsets or delays or anything. So, ntpd loads the network with 8 times as many queries as it actually uses. If there are large variations in the delays on the two paths, this is probably as good as you can do. If there are not, this is a very wasteful procedure and means you throw away data that you could have used to better control your clock. Chrony remembers up to 64 past offsets and rates, and uses those in a linear regression to get the best estimate of the current offset and rate. It changes how long a memory it uses by estimating whether or not the a linear fit is a good approximation to that data series, and throws away old measurements until it does feel that a linear fit is a good estimate. It does not try to do assymmetric path corrections which might be a detraction of chrony in some situations (my measurements indicate it is not on the systems I have looked at at least, but there may be situations in which it might help). BEcause of the clock selection algorithm, and the conservative choice for alpha, ntpd responds very slowly to changes in the rates due to temperature changes for example. chrony is much faster, and I believe this is one of the reasons for the better control that chrony gives.
Re: [ntp:questions] Red Hat vote for chrony
On 2014-12-05, David Woolley david@ex.djwhome.demon.invalid wrote: On 05/12/14 08:42, William Unruh wrote: Nope. ntp changes the rate of the local clock to correct offsets. That is all it does. It does not make the rate correct, and then the offset. Not true. It applies a correction for the offset, such that if that were the only error, it would be eliminated exponentially with a time constant that is related to the poll rate. I believe it expects to update this correction well before it is fully applied. ??? Not sure what you mean by not true. ntpd corrects offsets by changing the rate. ntp corrects rate changes by seeing the cumulative offset changes and changes the rate to correct the offsets. At the same time, it applies a correction to the long term estimate of the frequency, with a much longer time constant. much more quickly than ntpd does, for the same poll. Remember that ntpd throws out 85% of the measurements it makes, in order to try (poorly) to If the round trip variability was sufficiently low that this pre-filter were not needed, any variations in the samples it is ignoring would be largely suppressed by the loop filters, which have time constants that are significantly longer than even eight times the poll interval. ntpd is heavily oversampling, which means that the min and maxpols quoted for chrony imply an even higher excess rate. Those quotes were simply wrong. to quote from the chrony docs 3.4.2 Typical configuration files. -- To illustrate how a dial-up home computer might be configured, example configuration files are shown in this section. For the '/etc/chrony.conf' file, the following can be used as an example. server 0.pool.ntp.org minpoll 5 maxpoll 10 maxdelay 0.4 offline server 1.pool.ntp.org minpoll 5 maxpoll 10 maxdelay 0.4 offline server 2.pool.ntp.org minpoll 5 maxpoll 10 maxdelay 0.4 offline ... Does that look any different from what ntpd recommends? Also from the same document (chrony.txt in the docs that come with chrony) 'minpoll' Although 'chronyd' will trim the rate at which it samples the server during normal operation, the user may wish to constrain the minimum polling interval. This is always defined as a power of 2, so tt/minpoll 5/ would mean that the polling interval cannot drop below 32 seconds. The default is 6 (64 seconds). 'maxpoll' In a similar way, the user may wish to constrain the maximum polling interval. Again this is specified as a power of 2, so tt/maxpoll 9/ indicates that the polling interval must stay at or below 512 seconds. The default is 10 (1024 seconds). Actually not true. How do you think the standards of the various contries determine the accuracy of their clocks. They have no better time standard to compare them with. And yet they confidently will quote accuracy figures for their clocks. Study that. I presume that whatever time transfer mechanism they use has well characterised errors. NTP operates in an environment where the errors are not well characterised. Not true. Remember that they are operating down at the sub nanosecond level, and if you think transfer even across the room is well characterized at that level, But anyway that comment of mine was not about transfer errors. It was a reply tothe statement that the only way of characterizing the uncertainty of clocks is to have one that keeps better time than the clock one is looking at. But to get back to the chrony vs ntpd: chrony operates in an environment where the errors are not well characterised as well. IF you are in an environment where there are large (factors of 2) changes in the travel times in each direction, the clock filter of ntpd might give better results. It is something that has not AFAIK been tested. I am not sure what the answer would be. HOwever, that is not true of most situations in which ntpd or chrony are used. The dominant error is temperature fluctuations in which the clock frequency wanders around. ntpd handles that badly. chrony handles it better. So do you use a system which maybe protects you better in unusual situations, or one wich handles the typical case better? chrony could be altered to try to estimate whether or not one is in a case where there are large fluctuations in the travel times giving fluctuating assymetries in the travel time, and if so apply something like the ntpd clock filter. But it would have to be pretty bad I think before it became necessary. It would be a good master's project for someone to do the comparison. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Dec 5, 2014, at 11:47 AM, Paul tik-...@bodosom.net wrote: On Fri, Dec 5, 2014 at 9:37 AM, Charles Swiger cswi...@mac.com wrote: I also make sure that my timeservers are running in temperature-controlled environments so that such daily drifts you mention are minimized. I'm starting to think that people answering questions are unsure of the real question so they make a number of assumptions. If you care about sub-millisecond time then you need to say that and the question should be answered in that context. Well, we do have time enthusiasts around who like to achieve the best precision they can, regardless of whether there is a specific business justification or not. :-) I suspect most of the questions here refer to sub-second accuracy and most of the elaboration is unneeded. True; if you just want accuracy to the nearest second, you don't need to do anything elaborate to achieve that. Absent any other requirements, I think it reasonable for folks to target ~1 millisecond level of accuracy, which is quite doable on many platforms, even using remote timeservers over the WAN. If all your external clocks fail I suspect the typical user can depend on the disciplined virtual clock for days. For real hardware, sure-- once the intrinsic frequency drift has been setup, you can free-run for days into weeks without drifting too far. Cell phone towers (especially CDMA) are a decent example of such fault-tolerant systems. For almost all of human history, the sun or the fixed celestial heavens have provided the most accurate time reference available. Even today, we add (or subtract, in theory) leap seconds in order to keep UTC and UT1 aligned to better than a second courtesy of IERS. Yes, the USNO, CERN, and so forth now do have sufficiently high quality atomic clocks which have better timekeeping precision than celestial observations. I think there's some confusion here. Search for BIPM paper clock or read http://www.ggos-portal.org/lang_en/GGOS-Portal/EN/Topics/Services/BIPM/BIPM.html What confusion? Certainly it's a decent paper to read Such a point is orthogonal to the notion of how to measure a local clock I think this is an interesting question. How does one get high resolution measurements of the error in the virtual clock maintained with NTP (or Chrony)? I thought it was done with purpose built systems. Yes, you need to compare timestamps using purpose-built systems like a TCXO, Cesium, or Rubidium clock connected ideally via fast-interrupt driven parallel, serial, or network port which hopefully also provides hardware timestamping to minimize the processing latency. I don't expect a random version of Linux on generic hardware to be able to maintain the clock at nanosecond scale. True. I don't expect any version of Linux to perform at nanosecond scale, but that has as much to do with kernel bugs and compromises in timekeeping that particular OS has chosen. Even back in 2002 with very inexpensive commodity hardware, FreeBSD was able to achieve accuracy measured to ~260 nanoseconds: http://phk.freebsd.dk/soekris/pps/ ...per a Rb-based atomic clock. This is the sort of analysis I want to see for chrony. Regards, -- -Chuck ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Dec 5, 2014, at 11:53 AM, William Unruh un...@invalid.ca wrote: [ ... ] It simply alters the rate at any time so as to decrease the offset, and it does this measurement by measurement. It has no memory. This is obviously false. What do you think /etc/ntp.drift is? It is the offset from the standard rate of the clock. That memory is never used except on bootup. ntpd has to know how much to alter the drift. Ah, so you acknowledge that your original statement was wrong. Just like your claim whether the chrony docs recommend using maxpoll=4 across the network to a LAN timesource or not was wrong? Just like your claim about whether ntpd cares about figuring out the local clock frequency or whether it only chases the offset was wrong? Regards, -- -Chuck ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Fri, Dec 5, 2014 at 3:35 PM, Charles Swiger cswi...@mac.com wrote: Well, we do have time enthusiasts around who like to achieve the best precision they can, regardless of whether there is a specific business justification or not. :-) Sure but that doesn't help someone that just wants a simple (-minded) configuration to keep a few clocks in sync. If all your external clocks fail I suspect the typical user can depend on the disciplined virtual clock for days. For real hardware, sure-- once the intrinsic frequency drift has been setup, you can free-run for days into weeks without drifting too far. Cell phone towers (especially CDMA) are a decent example of such fault-tolerant systems. c While perhaps strictly correct I was talking about the common crystal in a typical computer not the Cesium, Rubidum or OCXO in a cell site and there's really no basis for comparison. What confusion? Certainly it's a decent paper to read I misquoted. This Without measuring the local clock against some other clock or oscillator suggests comparing a clock to a better frequency reference but BIPM creates a virtual clock (with better Allen deviation) and everyone steers toward that. Perhaps you meant something else. Yes, you need to compare timestamps using purpose-built systems like a TCXO, Cesium, or Rubidium clock That wasn't my point. You need a purpose built NTP server to expose its virtual clock for comparison to an external frequency reference. Of course you need a purpose built reference but only in the sense that you'd use real counter rather than one in a voltmeter. Even back in 2002 with very inexpensive commodity hardware, FreeBSD was able to achieve accuracy measured to ~260 nanoseconds: H. So phk uses a $1,500 rubidium standard as a system oscillator and you call it inexpensive and commodity. He also ran a particular install of BSD and a non-standard NTP. All of those are what I was referring to when I said purpose built system to measure the variance of an NTP disciplined virtual oscillator. By the way high resolution low-latency counters in computers have become commodity items. The software to use them -- not so much. It might be nice to conduct a similar experiment with Chrony but it's all pointless. As Bill suggests you want to measure typical performance in typical environments. That's the bit I said was interesting and I don't think the published numbers make it clear what's better. Frankly I suspect even that is pointless. If someone asked me how to do NTP on the cheap I'd say buy or build some number of Laureline-like PLLs in a box with a NTP packet emulator or some NTP servers off Ebay and run SNTP/OpenNTP on the clients. If you have a larger budget then buy Meinberg, Microsemi et. al. new. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Dec 5, 2014, at 5:55 PM, Paul tik-...@bodosom.net wrote: [ ... ] Even back in 2002 with very inexpensive commodity hardware, FreeBSD was able to achieve accuracy measured to ~260 nanoseconds: H. So phk uses a $1,500 rubidium standard as a system oscillator and you call it inexpensive and commodity. No, he used a $1500 rubidium clock to accurately measure the timekeeping quality of a $220 Soekris computer, and concluded: I have earlier complained that no good and cheap hardware were available which could timestamp a PPS signal reliably and precisely but now the Soekris computer has proven that it does indeed deliver just that: With a pricetag of approx USD220 (single unit including enclosure) this is the best hardware you can find for a stratum 1 NTP server. If you wanted to drive such hardware via a ~$40 GPS puck, you'd probably see an accuracy of around a microsecond, perhaps a bit worse depending on the timekeeping accuracy which the GPS puck provides. That was also the level of accuracy I was seeing from generic Intel hardware running FreeBSD as a stratum 1 with a GPS source. I've used a digital frequency counter which had an onboard TCXO (or possibly a DTCXO) for measuring. Although the frequency counter supported receiving higher-quality PPS timing from an external atomic clock, I've never had a Cs or Rb source, so I won't claim to have measured sub-microsecond accuracy with it. He also ran a particular install of BSD and a non-standard NTP. I believe he ran FreeBSD 4.x and likely the ntpd from ports. Regards, -- -Chuck ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Dec 5, 2014, at 5:55 PM, Paul tik-...@bodosom.net wrote: [ ... ] Even back in 2002 with very inexpensive commodity hardware, FreeBSD was able to achieve accuracy measured to ~260 nanoseconds: H. So phk uses a $1,500 rubidium standard as a system oscillator and you call it inexpensive and commodity. No, he used a $1500 rubidium clock to accurately measure the timekeeping quality of a $220 Soekris computer, and concluded: I have earlier complained that no good and cheap hardware were available which could timestamp a PPS signal reliably and precisely but now the Soekris computer has proven that it does indeed deliver just that: With a pricetag of approx USD220 (single unit including enclosure) this is the best hardware you can find for a stratum 1 NTP server. If you wanted to drive such hardware via a ~$40 GPS puck, you'd probably see an accuracy of around a microsecond, perhaps a bit worse depending on the timekeeping accuracy which the GPS puck provides. That was also the level of accuracy I was seeing from generic Intel hardware running FreeBSD as a stratum 1 with a GPS source. I've used a digital frequency counter which had an onboard TCXO (or possibly a DTCXO) for measuring. Although the frequency counter supported receiving higher-quality PPS timing from an external atomic clock, I've never had a Cs or Rb source, so I won't claim to have measured sub-microsecond accuracy with it. He also ran a particular install of BSD and a non-standard NTP. I believe he ran FreeBSD 4.x and likely the ntpd from ports. Regards, -- -Chuck ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Dec 5, 2014, at 5:55 PM, Paul tik-...@bodosom.net wrote: [ ... ] Even back in 2002 with very inexpensive commodity hardware, FreeBSD was able to achieve accuracy measured to ~260 nanoseconds: H. So phk uses a $1,500 rubidium standard as a system oscillator and you call it inexpensive and commodity. No, he used a $1500 rubidium clock to accurately measure the timekeeping quality of a $220 Soekris computer, and concluded: I have earlier complained that no good and cheap hardware were available which could timestamp a PPS signal reliably and precisely but now the Soekris computer has proven that it does indeed deliver just that: With a pricetag of approx USD220 (single unit including enclosure) this is the best hardware you can find for a stratum 1 NTP server. If you wanted to drive such hardware via a ~$40 GPS puck, you'd probably see an accuracy of around a microsecond, perhaps a bit worse depending on the timekeeping accuracy which the GPS puck provides. That was also the level of accuracy I was seeing from generic Intel hardware running FreeBSD as a stratum 1 with a GPS source. I've used a digital frequency counter which had an onboard TCXO (or possibly a DTCXO) for measuring. Although the frequency counter supported receiving higher-quality PPS timing from an external atomic clock, I've never had a Cs or Rb source, so I won't claim to have measured sub-microsecond accuracy with it. He also ran a particular install of BSD and a non-standard NTP. I believe he ran FreeBSD 4.x and likely the ntpd from ports. Regards, -- -Chuck ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On 2014-12-05, Charles Swiger cswi...@mac.com wrote: On Dec 5, 2014, at 11:53 AM, William Unruh un...@invalid.ca wrote: [ ... ] It simply alters the rate at any time so as to decrease the offset, and it does this measurement by measurement. It has no memory. This is obviously false. What do you think /etc/ntp.drift is? It is the offset from the standard rate of the clock. That memory is never used except on bootup. ntpd has to know how much to alter the drift. Ah, so you acknowledge that your original statement was wrong. Just like your claim whether the chrony docs recommend using maxpoll=4 across the network to a LAN timesource or not was wrong? I have given you the quotes from the chrony docs. Where do you get your statement from? Just like your claim about whether ntpd cares about figuring out the local clock frequency or whether it only chases the offset was wrong? ntpd does not care about figuring out the local clock frequency. It simply adjusts the frequency to minimize the offset. That is all it does. Yes, the computer (or if you have a kernel which does not allow you to adjust the computer's frequency ntpd) does have a frequency. But no memory of that is kept. ntpd does not know what the frequency was before it currently adjusted the frequency, and especially not what it was 5 adjustments ago. Regards, ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On 2014-12-04, David Woolley david@ex.djwhome.demon.invalid wrote: I've been reading the migration guidance document for RHEL 7 and it seems that Chrony has replaced ntpd as their default NTP based time synchonisation package. https://access.redhat.com/documentation/en-US/Red_Hat_Enterprise_Linux/7/pdf/Migration_Planning_Guide/Red_Hat_Enterprise_Linux-7-Migration_Planning_Guide-en-US.pdf They quote a list of benefits, but I think that is just quoting the upstream claims, and it took them a long time to disable the local clock, so they are not exactly experts. Actually Miroslav Lichvar IS an expert. He is the chrony maintainer, has done a lot of testing comparing chrony to ntpd ( which showed that chrony controlled the clock a factor of 2 to 20 times better than ntpd did), and is with Redhat. ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions
Re: [ntp:questions] Red Hat vote for chrony
On Dec 4, 2014, at 7:00 PM, William Unruh un...@invalid.ca wrote: [ ... ] Actually Miroslav Lichvar IS an expert. He is the chrony maintainer, has done a lot of testing comparing chrony to ntpd ( which showed that chrony controlled the clock a factor of 2 to 20 times better than ntpd did), and is with Redhat. The data I've seen for chrony suggests it handles broken clocks such as commonly found in VMs better than ntpd does. The tradeoff is that chrony prioritizes chasing the reference time over first trying to ensure that the local clock frequency is stable, whereas ntpd really wants to make sure that the local clock counts 3600 seconds in each hour of wall-clock time and then worries about slewing the local time to match up with the reference time. It's informative to note that the chrony docs (section 5.3.4) recommend using minpoll=2 and maxpoll=4! With those settings chrony will send 225 polls per hour, versus 3.5 polls per hour for ntpd with its maxpoll=10. Assuming arguendo the claim of a factor of 20 times better is true, I still don't care to pay the price of a factor of 64 times more network polls. Furthermore-- unfortunately-- I have yet to see data on the accuracy of chrony measured against high-quality TCXO or Rb/Cs reference clocks, such as the PRS-10 that PHK used: http://www.thinksrs.com/products/PRS10.htm ...the current version of which claims to have a +/- 10 ns accuracy for the PPS signal. Instead, most of the data I've seen provided for chrony has involved comparing local clock timestamps to the reference timesource or to some other network timesource, without detailed information as to the accuracy of those references. Of course you're not going to see much delta between the local clock and the reference that you're polling every 16 seconds. Without measuring the local clock against some other clock or oscillator which is known to be accurate to sub-microsecond levels, one doesn't have the data needed to draw conclusions about the actual timekeeping precision. Regards, -- -Chuck ___ questions mailing list questions@lists.ntp.org http://lists.ntp.org/listinfo/questions