Pendulum clock runs *fast* in hot weather

[newshound]

Here's a better one

0.6 seconds over 100 days, and that was a pendulum clock to Harrison's design (though not, of course, at sea on a sailing ship).

[NY]

Yes but it was probably a far more elaborate mechanism with far more checks and counter checks that a normal clock mechanism. To make something that keeps consistent time, in a range of temperatures and on board a heaving ship, was a fantastic achievement. And it must have taken ages to adjust the clock, little by little, until it was keeping the correct time, even if once it *had* been set, it remained accurate. I presume it was a rotating balance wheel as opposed to a pendulum, so it would be a matter of tweaking the length of the spring. And what would they have used as a reference against which to measure how far fast or slow the Harrison watch was, while adjusting it. I suppose for a fixed location, the movement of the stars is known so you can tell the time fairly accurately using those and compare the watch against that.

Come to think of it, I wonder why my PC clock is so bad. Quartz watches keep better time and don't need much adjustment back to correct time. I wonder if when the PC is on, it uses some other time source than the quartz crystal.

[newshound]

PC clocks were never particularly accurate because they didn't need to be. Doesn't everything get time from the internet these days?

[Bob Eager]

We have one clockwork clock. All the rest get their time via NTP, or from Anthorn.

[Jethro_uk]

Which, given the nature of TCP/IP is quite a feat.

Trouble is you need *really* accurate timestamps now, if you are dabbling in cryptobollocks ^H^H^H^H^H^H^H^H. Which can put a hell of a strain on your CPU, as your PC desperately tries to sync with a never ending blockchain.

[NY]

A clock only needs to be as accurate as the gain or loss that you can tolerate in the interval until it is next synchronised with a master time source.

Windows defaults to syncing once a week. After I found my PCs drifting by several minutes, I changed that to once every day. It's a shame that it requires a registry change, and isn't changeable from a menu built into Windows.

As I understand it, the Raspberry Pi doesn't have a battery-backed real time clock, and has to set itself afresh every time it is booted and can talk to an NTP server. I presume it resyncs periodically while it is running.

I'm cautious of clocks that have no free-running capability, which depend entirely on a radio source. My wife bought a Rugby clock/radio years ago and towards the end it started displaying bizarre times, probably if it lost the radio signal, because it couldn't keep going (even if with a slight error) until the next time it got a valid time. Not good when it's the alarm clock that wakes you to go to work :-(

[Harry Bloomfield]

T i m submitted this idea :

Less drag, usually means the pendulum can swing higher, the higher it swings, the slower it ticks.

[T i m]

Wouldn't all that be down to the OS running on it, so not the RPi itself?

On one ESP32 project I built it makes a WiFi connection, connects to an NTP server, logs the values of a temperature sensor to an SD card and then goes to sleep for n time.

On my Arduino projects I have an external RTC.

Cheers, T i m

[Harry Bloomfield]

NY explained on 03/08/2018 :

I would expect much better than that, from a half decent mechanical watch.

Again it depends upon the quality of the unit. The none adjustable ones, yes +/- 5 seconds. I can usually adjust a decent one to +/- 1 second per week kept on the wrist.

Like some clocks, those are the very least accurate usually with no way to trim them for best accuracy, besides there location subjects them to wide temperature variations.

The most accurate crystal based clocks, have the crystal and oscillator mounted in an oven or a double oven system. I have one which usually manages 0.1 second per month.

I have a cheap Sekonda quartz, which manages around a surprising one second per month on the wrist.

Most consistently accurate, for year in year out, are radio controlled watches and clocks or ones synched via mobile mast/ internet.

[The Natural Philosopher]

Of course it doees. NTP on the internet> PROBABLY accurate to a ms or so.

[Harry Bloomfield]

NY used his keyboard to write :

Microsoft syncs them to the Internet once per day.

[Jethro_uk]

Not good enough for syncing a blockchain ....

[T i m]

But that's not a lighter / faster bit or the issue is it?

(The idea is) It's less drag on the things that regulate the thing that regulates the time. ;-)

Do you have a more plausible reason?

Cheers, T i m

[NY]

Ah, sorry, you misunderstood me. I was meaning for its free-running, in between fairly rare resetting by an NTP source. I seem to remember some bollocks about the RTC on a PC only using its quartz crystal to keep time when the PC was off, and that it used another way of keeping time (using the CPU) while the PC was booted to an operating system. But I can't remember enough details to google it.

And then of course, while it is booted, it can synchronise to a master time source to kick it back to correct time if the local tick (whether crystal or something else) has drifted a bit.

Windows' default for NTP is as infrequent as once a week. I found the registry tweak to change this to every 24 hours.

I've just checked with time.is and the clock is current 0.7 sec ahead, having last synced at 22:15 last night (for once, the Date and Time | Internet Time tab actually tells me when it last synced, usually that info is not given for some reason).

[NY]

I'm sure by default it's once a week. I remember changing it to once a day when I found that my clock was sometimes as much as a couple of minutes wrong when it was syncing once a week.

[The Natural Philosopher]

That certainly was correct the last time I was involved - free running oscillator taht is realtively ribbish at timekeeing.

IIRC linux is almost every minute...not sure

"Generally we can also call it as polling interval and minimum time is

64 sec and maximum time 1024 sec , but you can still change it as you want by doing changes at /etc/ntp.conf."

[Harry Bloomfield]

T i m formulated on Friday :

No, just expressing my thoughts on the subject from knowing how mechanical time-pieces generally behave. My best guess would be that the pendulum does have some compensation built in and the compensation is over compensating when the weather heats up.

[Harry Bloomfield]

NY wrote on 03/08/2018 :

I have just checked and W10's default is once per week, with no means to change it - earlier versions did it once per day. I use a utility which sets my PC once every hour NetTime, because I am fixing a few of my old watches up at the moment and I need accurate references.

[Robin]

Ahem, I think you may temporarily have overlooked that a - if not the - point about using a pendulum for timekeeping is that the period of oscillation is independent of the amplitude.

[Harry Bloomfield]

It happens that Jethro_uk formulated :

The clock setting utils, check the average path delay to the time servers, then adjust the time suit.

give a constant accurate on screen time reference, compared against your PC's clock.