The standard thing to do is to build another one and compare the two: the stability of a single clock should be sqrt(2) smaller. If you have three clocks, then you can determine the stability of each clock. This doesn't tell you about systematic errors but physics comes to the rescue there because the influences that cause systematic frequency errors (magnetic fields, collisions, black body radiation...) can be characterised and controlled.
I know you were just joking but... it's not a clock. It's a frequency reference. A clock needs to run continuously and the state of the art 'clocks' like lattice clocks are not reliable enough to do this yet. Very good frequency references can be used to correct the frequency of less accurate clocks and this is what happens in the computation of Coordinated Universal Time. Time of day is computed from 400 or so continuously running hydrogen masers and caesium beam standards and then a very small correction is made using 'one second' as calculated from half a dozen or so extremely good frequency standards like caesium fountains.
Not quite. How do you set time of day? It's only a frequency reference, not a time reference. For time references, there's really only GNSS, dialup time services like ACTS and the various radio services (and maybe Iridium). The value of the CSAC, synchronised with GNSS, is that it will give you better holdover than the very ordinary crystal oscillator in your PC if GNSS is unavailable. As for the frequency offset, the CSACs 'age,' that is their frequency offset increases with time, so that the time offset increases quadratically.
Back in the 1980's there was a reanalysis of some old gravity measurements made by Roland von Eotvos which suggested that gravity might have a short-range, composition-dependent component, a "fifth force". This inspired a number of experiments, with some positives and some negative results. Eventually, the positive results were all explained and the fifth force went away.
Coincidentally, in regard to this recent research, one of the hard to explain positive results also came out of UC Irvine.
And back in the 1990's we had BMRT (a free renderman clone); until they came and paid/threatened the guy to stop making the free clone available.
Sorta. Larry Gritz, the author of BMRT, went to work for Pixar and then left to start his own company, Exluna, whose main product was a Renderman competitor called Entropy. Unfortunately Pixar's lawyers jumped on Exluna and Exluna was vaporised. BMRT and Entropy were no longer available after this. Larry Gritz went to work for Nvidia after that on a GPU-accelerated renderer, I think.
They are the official timekeepers for the US, along with the U.S. Naval Observatory (which also operates the timescale that GPS satellite clocks are steered to) but they are not timekeepers for the world.
The international standard for time is UTC, a 'paper' clock which is the average of atomic clocks from all around the world.
Standard OS clocks only tick at about 100hz, so you're always out by an average of 5ms anyways.
Nope. Although the system interrupt is only between a few hundred Hz and a kHz, other, faster counters are used to interpolate between these ticks. So on Linux, eg the Time Stamp Counter in the CPU can be used to improve the timestamp resolution to a microsecond, or even nanoseconds, with the nanokernel patch (which is standard in the BSDs, I think).
In my experience of operating a network of geographically-dispersed stratum 1 NTP servers, there is frequently asymmetry at the 1 to 2 s level and occasionally worse. An NTP implementation like ntpd filters out these outliers but the simple protocol you are suggesting would not.
PTP cannot account for network asymmetries any more than NTP can. It can only guarantee symmetric paths when all the hardware between two endpoints is PTP-capable, meaning that each boundary device has to implement a PTP clock.
In the end, it seems silly not to synchronise device clocks to a universal reference . There are many local applications which require a timestamp that must be compared with a time stamp on another device at some time down the line.
"You should try to keep up.:)" I think the poster can be forgiven for not knowing about an alpha-release NTP client that only works on *nix at the moment (and was only released 3 months ago).
"network latency adjustment is automatic" - I don't understand this statement. If you are only taking two time stamps - client transmit and server receive - then you have no information about the network latency.
>>Nope. I don't know anyone over 50 who knows how to build or repair a steam engine. Amusingly, I know three, and one of them is under 50. But my sample is probably biased: the people I work with are all physicists or engineers.
I just read a SF story that used an idea like this - that analogous to the time dilation experienced as you fell into a 'normal' black hole, you would see spatial dilation as you fell into a black hole 'in time'. the idea that the universe is inside a black hole has been bandied about but I can't find a reference for the 'temporal black hole' idea ( I feel sure that the story I read was based on a scientific paper).
"They may not be able to illiterate why "... Darn it! What is that word? It's on the tip of my tongue ! Why can't I remember the word that describes my inability to summon the right word just when I need it?. Oh.... It's 'articulate'! Dementia denied once more.
We had a high resolution, full-body scanner at work that was being used to build a database of body shapes. I volunteered but was rather dismayed when I loaded the 3-d model to see what shape I really am...
One more thought before I go off to sleep... The spacecraft is the source of the magnetic field so that means the field lines have to terminate on it. Which means hot plasma is continuously blasting into you.
The paper is a one pager of introductory plasma physics. It isn't a serious calculation and it wasn't meant to be. Anyway...
Their model is as follows. A plasma will reflect all electromagnetic radiation below a certain frequency, determined by its density. The plasma exerts a pressure like a gas and they then assume confinement of the plasma with a magnetic field, balancing the plasma pressure with the 'pressure' that a magnetic field exerts on charged particles. They then say that we can make magnetic fields in the range up to 100 T and working back, estimate the plasma frequency, which turns out to be in the UV. So great, you can deflect lasers into the UV with a modest confining field.
You need to look at some of the other numbers though. First, what sort of plasma density do you need to reflect UV ? The answer is something like 10^28 per cubic m. This is enormous - fusion plasmas are about a million times less dense). It's getting close to solid state density eg if a solid has atoms 0.2 nm apart this is 10^29 atoms per cubic m. That is not going to be easy.... The other problem is that at such a high density, the collision frequency is very high so that a magnetic field is not very effective at producing confinement. Probably useless in fact.
The other thing to look at is the required plasma temperature. They assume a temperature of 1000 K, Unfortunately, the density of a plasma at 1000 K at thermal equilibrium is extremely low unless the background pressure is huge. So it has to be a lot hotter, in particular, comparable with the ionization energy which is roughly 100 000 K. And really, we need a fully ionized plasma because the magnetic field is not going to confine the neutral gas that we are using to make the plasma so that means we need a 100 000 K plasma. This means that the required magnetic field goes up by a factor of 10.
Would somebody else like to estimate how much power you need to dump into the plasma ?
Typically, there is no single clock that is "the reference". The international time standard, UTC, is an average of about 400 or so atomic clocks from all around the world. There are a number of caesium fountains currently contributing to UTC. Even in a single laboratory, where there a number of clocks, these clocks will usually be averaged and one clock then adjusted to keep to this average. There are various practical reasons for this. If you have a number of similar clocks, the average will have a more stable frequency than a single clock. Some clocks are better at very short averaging times, others better for the long term so you can make an average that takes this into account. State-of-the-art clocks seldom work continuously so you need something else as a flywheel between operating times.
The main use for very good clocks is as frequency references, or to measure the time interval between two events (in which case any offset cancels out) rather than time of day references. Sub-nanosecond synchronisation of distributed systems is only used in some very specialised scientific applications like Very Long Baseline Interferometric telescopes.
You can sync it with existing clocks to an accuracy of perhaps 500 picoseconds using well-established techniques like GPS-carrier phase and Two-way Satellite Time-Transfer.
Not sure if the parent is simply being funny, but... No atomic clock is ever 'reset' because of leap seconds. All they produce is a one-pulse-per-second 'tick'. The labels on those ticks are completely arbitrary. When a leap second occurs, you just change the labels...
The receiver might cost $60 an antenna suitable for permanent outdoor installation might cost the same, but running cable to the roof, assuming that's possible, could easily cost a few thousand dollars to do properly. You'll want lightning protection too, if you're running the antenna cable down in to your server racks.
But for nearly everyone, as others point out, the 10 ms or so accuracy you will get from a nearby NTP server, is more than enough.
In short, "counting vibrations" is a poor description of what's going on. In the case of a caesium fountain, what you're doing is driving a resonant transition in the atom with external microwaves. When the microwaves are tuned to the transition, the Cs atom changes its state, which can be detected. To make the microwaves, you usually start with something like a very good 5 MHz crystal oscillator and multiply it up (plus add an offset, also derived from the crystal) to get the transition frequency ( for the transition in Cs which defines what one second is, this is an exact frequency) . You then adjust the frequency of the crystal (this is done electrically) until you're centred on the resonance in the Cs atom. So then you can just electrically count the oscillations of the crystal and 5000000 of these will be your definition of what one second is.
I use LabVIEW for FPGA programming, with other LabVIEW code running on a real-time controller interfaced to the FPGA, with top-level control, logging etc on a PC. LabVIEW is nice for the FPGA because the metaphor suits the parallelism of FPGAs. Part of why it works too is that you're only dealing with a very limited subset of LabVIEW's functionality and you can't do complicated things, at least if you want to clock at high speed. But the code running on the real-time controller and PC... yuck. I'm an experienced C-coder and LabVIEW can be painful when algorithms eg for a servo, get complex. The standard joke "LabVIEW makes easy things easy" has plenty of truth in it. What it means is that once you get beyond a certain level of complexity, LabVIEW doesn't offer any advantage over a text-based language.
From the article "Gary Shu, XYZprinting's market development division senior manager, said the 3D printer can quickly create objects that users may need in their homes, such as a plastic cup or a plastic spoon.". I hope he comes up with a few better ideas than that.
Actually, a 3D printer would be useful to me for hobby projects like cosplay props, although probably a bit expensive. But around the house ? I look around for things completely made out of plastic that it would be practical to print if they broke or I needed another one but it's a struggle.
I suppose what all of these 3D printer manufacturers want to convince themselves and their investors is that there is a mass market for their product. The cheap printers still look very much like a hobbyist tool to me though.
Slashdot Mirror
The standard thing to do is to build another one and compare the two: the stability of a single clock should be sqrt(2) smaller. If you have three clocks, then you can determine the stability of each clock. This doesn't tell you about systematic errors but physics comes to the rescue there because the influences that cause systematic frequency errors (magnetic fields, collisions, black body radiation ...) can be characterised and controlled.
I know you were just joking but ... it's not a clock. It's a frequency reference. A clock needs to run continuously and the state of the art 'clocks' like lattice clocks are not reliable enough to do this yet. Very good frequency references can be used to correct the frequency of less accurate clocks and this is what happens in the computation of Coordinated Universal Time. Time of day is computed from 400 or so continuously running hydrogen masers and caesium beam standards and then a very small correction is made using 'one second' as calculated from half a dozen or so extremely good frequency standards like caesium fountains.
Not quite. How do you set time of day? It's only a frequency reference, not a time reference. For time references, there's really only GNSS, dialup time services like ACTS and the various radio services (and maybe Iridium). The value of the CSAC, synchronised with GNSS, is that it will give you better holdover than the very ordinary crystal oscillator in your PC if GNSS is unavailable.
As for the frequency offset, the CSACs 'age,' that is their frequency offset increases with time, so that the time offset increases quadratically.
Back in the 1980's there was a reanalysis of some old gravity measurements made by Roland von Eotvos which suggested that gravity might have a short-range, composition-dependent component, a "fifth force". This inspired a number of experiments, with some positives and some negative results. Eventually, the positive results were all explained and the fifth force went away.
Coincidentally, in regard to this recent research, one of the hard to explain positive results also came out of UC Irvine.
Indeed not. Top pay for a scientific/technical position in the Commonwealth public service is about $130K.
And back in the 1990's we had BMRT (a free renderman clone); until they came and paid/threatened the guy to stop making the free clone available.
Sorta. Larry Gritz, the author of BMRT, went to work for Pixar and then left to start his own company, Exluna, whose main product was a Renderman competitor called Entropy. Unfortunately Pixar's lawyers jumped on Exluna and Exluna was vaporised. BMRT and Entropy were no longer available after this. Larry Gritz went to work for Nvidia after that on a GPU-accelerated renderer, I think.
They are the official timekeepers for the US, along with the U.S. Naval Observatory (which also operates the timescale that GPS satellite clocks are steered to) but they are not timekeepers for the world.
The international standard for time is UTC, a 'paper' clock which is the average of atomic clocks from all around the world.
Standard OS clocks only tick at about 100hz, so you're always out by an average of 5ms anyways.
Nope. Although the system interrupt is only between a few hundred Hz and a kHz, other, faster counters are used to interpolate between these ticks. So on Linux, eg the Time Stamp Counter in the CPU can be used to improve the timestamp resolution to a microsecond, or even nanoseconds, with the nanokernel patch (which is standard in the BSDs, I think).
In my experience of operating a network of geographically-dispersed stratum 1 NTP servers, there is frequently asymmetry at the 1 to 2 s level and occasionally worse. An NTP implementation like ntpd filters out these outliers but the simple protocol you are suggesting would not.
PTP cannot account for network asymmetries any more than NTP can. It can only guarantee symmetric paths when all the hardware between two endpoints is PTP-capable, meaning that each boundary device has to implement a PTP clock.
In the end, it seems silly not to synchronise device clocks to a universal reference . There are many local applications which require a timestamp that must be compared with a time stamp on another device at some time down the line.
"You should try to keep up. :)"
I think the poster can be forgiven for not knowing about an alpha-release NTP client that only works on *nix at the moment (and was only released 3 months ago).
"network latency adjustment is automatic" - I don't understand this statement.
If you are only taking two time stamps - client transmit and server receive - then you have no information about the network latency.
>>Nope. I don't know anyone over 50 who knows how to build or repair a steam engine.
Amusingly, I know three, and one of them is under 50. But my sample is probably biased: the people I work with are all physicists or engineers.
I just read a SF story that used an idea like this - that analogous to the time dilation experienced as you fell into a 'normal' black hole, you would see spatial dilation as you fell into a black hole 'in time'. the idea that the universe is inside a black hole has been bandied about but I can't find a reference for the 'temporal black hole' idea ( I feel sure that the story I read was based on a scientific paper).
"They may not be able to illiterate why " ... Darn it! What is that word? It's on the tip of my tongue ! Why can't I remember the word that describes my inability to summon the right word just when I need it?. Oh .... It's 'articulate'! Dementia denied once more.
We had a high resolution, full-body scanner at work that was being used to build a database of body shapes. I volunteered but was rather dismayed when I loaded the 3-d model to see what shape I really am ...
Two dollars is a lot - paperbacks I bought in the late 1970's were under a dollar so your theory about an inflated price does look plausible.
One more thought before I go off to sleep ...
The spacecraft is the source of the magnetic field so that means the field lines have to terminate on it. Which means hot plasma is continuously blasting into you.
The paper is a one pager of introductory plasma physics. It isn't a serious calculation and it wasn't meant to be. Anyway ...
Their model is as follows. A plasma will reflect all electromagnetic radiation below a certain frequency, determined by its density. The plasma exerts a pressure like a gas and they then assume confinement of the plasma with a magnetic field, balancing the plasma pressure with the 'pressure' that a magnetic field exerts on charged particles. They then say that we can make magnetic fields in the range up to 100 T and working back, estimate the plasma frequency, which turns out to be in the UV. So great, you can deflect lasers into the UV with a modest confining field.
You need to look at some of the other numbers though. .... The other problem is that at such a high density, the collision frequency is very high so that a magnetic field is not very effective at producing confinement. Probably useless in fact.
First, what sort of plasma density do you need to reflect UV ? The answer is something like 10^28 per cubic m. This is enormous - fusion plasmas are about a million times less dense). It's getting close to solid state density eg if a solid has atoms 0.2 nm apart this is 10^29 atoms per cubic m. That is not going to be easy
The other thing to look at is the required plasma temperature. They assume a temperature of 1000 K, Unfortunately, the density of a plasma at 1000 K at thermal equilibrium is extremely low unless the background pressure is huge. So it has to be a lot hotter, in particular, comparable with the ionization energy which is roughly 100 000 K. And really, we need a fully ionized plasma because the magnetic field is not going to confine the neutral gas that we are using to make the plasma so that means we need a 100 000 K plasma. This means that the required magnetic field goes up by a factor of 10.
Would somebody else like to estimate how much power you need to dump into the plasma ?
Typically, there is no single clock that is "the reference". The international time standard, UTC, is an average of about 400 or so atomic clocks from all around the world. There are a number of caesium fountains currently contributing to UTC. Even in a single laboratory, where there a number of clocks, these clocks will usually be averaged and one clock then adjusted to keep to this average. There are various practical reasons for this. If you have a number of similar clocks, the average will have a more stable frequency than a single clock. Some clocks are better at very short averaging times, others better for the long term so you can make an average that takes this into account. State-of-the-art clocks seldom work continuously so you need something else as a flywheel between operating times.
The main use for very good clocks is as frequency references, or to measure the time interval between two events (in which case any offset cancels out) rather than time of day references. Sub-nanosecond synchronisation of distributed systems is only used in some very specialised scientific applications like Very Long Baseline Interferometric telescopes.
You can sync it with existing clocks to an accuracy of perhaps 500 picoseconds using well-established techniques like GPS-carrier phase and Two-way Satellite Time-Transfer.
Not sure if the parent is simply being funny, but ... ...
No atomic clock is ever 'reset' because of leap seconds. All they produce is a one-pulse-per-second 'tick'. The labels on those ticks are completely arbitrary. When a leap second occurs, you just change the labels
The receiver might cost $60 an antenna suitable for permanent outdoor installation might cost the same, but running cable to the roof, assuming that's possible, could easily cost a few thousand dollars to do properly. You'll want lightning protection too, if you're running the antenna cable down in to your server racks.
But for nearly everyone, as others point out, the 10 ms or so accuracy you will get from a nearby NTP server, is more than enough.
In short, "counting vibrations" is a poor description of what's going on. In the case of a caesium fountain, what you're doing is driving a resonant transition in the atom with external microwaves. When the microwaves are tuned to the transition, the Cs atom changes its state, which can be detected. To make the microwaves, you usually start with something like a very good 5 MHz crystal oscillator and multiply it up (plus add an offset, also derived from the crystal) to get the transition frequency ( for the transition in Cs which defines what one second is, this is an exact frequency) . You then adjust the frequency of the crystal (this is done electrically) until you're centred on the resonance in the Cs atom. So then you can just electrically count the oscillations of the crystal and 5000000 of these will be your definition of what one second is.
I use LabVIEW for FPGA programming, with other LabVIEW code running on a real-time controller interfaced to the FPGA, with top-level control, logging etc on a PC. LabVIEW is nice for the FPGA because the metaphor suits the parallelism of FPGAs. Part of why it works too is that you're only dealing with a very limited subset of LabVIEW's functionality and you can't do complicated things, at least if you want to clock at high speed. ... yuck. I'm an experienced C-coder and LabVIEW can be painful when algorithms eg for a servo, get complex. The standard joke "LabVIEW makes easy things easy" has plenty of truth in it. What it means is that once you get beyond a certain level of complexity, LabVIEW doesn't offer any advantage over a text-based language.
But the code running on the real-time controller and PC
From the article "Gary Shu, XYZprinting's market development division senior manager, said the 3D printer can quickly create objects that users may need in their homes, such as a plastic cup or a plastic spoon.". I hope he comes up with a few better ideas than that.
Actually, a 3D printer would be useful to me for hobby projects like cosplay props, although probably a bit expensive. But around the house ? I look around for things completely made out of plastic that it would be practical to print if they broke or I needed another one but it's a struggle.
I suppose what all of these 3D printer manufacturers want to convince themselves and their investors is that there is a mass market for their product. The cheap printers still look very much like a hobbyist tool to me though.