[time] GPS Time Server Survey
Thu Apr 24 22:20:27 UTC 2008
Kipton Moravec wrote on 24-4-2008 23:19:
> On Thu, 2008-04-24 at 13:45 -0700, Chris Kuethe wrote:
>> On Thu, Apr 24, 2008 at 9:14 AM, Kipton Moravec <kip at kdream.com> wrote:
>>> I am designing a system with a GPS module, and started looking at all of
>>> the things it could do, including time.
>>> Since you folks are all interested in time, I was wondering which of two
>>> products would be of interest to you.
>>> 1. A GPS module with the ability to expand the serial port to more than
>>> 10 to allow you to use the NTP NEMA driver. The time pulse would be
>>> accurate to about 30 nS RMS. Retail cost about $100.
>> yes. although i'm wondering where you're going to get such a
>> non-crappy chip. Also, will it output raw measurements (CP/PR/D)?
>> While you're at it, could you mount this on a (mini)?pci card with a
>> dual-port UART - plug it in, it looks like 2 more com ports, but then
>> you can use one port as your time source and the other port for
>> "control" purposes.
> The problem with putting it in the PC is that the PC is most often not
> near the sky. GPS needs some view of the sky.
> You loose a lot of signal with a long antenna lead, that is why you
> rarely see more than a 15 foot antenna lead.
>>> 2. A time server with 2 Ethernet ports, with GPS time reference, to
>>> directly support between 50 and 100 users using GPS. Retail Cost about
>> could be fun - depends on what's in it.
> Just a 32-bit microcontroller with a good crystal and a GPS. All running
> linux. No keyboard, no monitor, just ssh to it over the Ethernet. Not
> that exciting.
>>> Both would be Stratum 1 by using the GPS (and later Galileo also)
>>> Any additional specifications? How much more would you pay for the
>>> additional specs?
>>> Or is this not interesting enough for you, that I would be wasting my
>> my home time server is a soekris net4801 driven by a garmin gps18;
>> pretty low-end, but internal diagnostics say that it's tracking within
>> 1 usec of the garmin's PPS output. that's tolerable, considering the
>> garmin is only specified to 1 usec accuracy, and the 4801 isn't really
>> a marvel of computing precision. And the fact that it's
>> ethernet-connected to the ass end of an ADSL line - there goes all my
>> precision anyway.
> This is 30 times more accurate than the Garmin, but the PC is no where
> near that accurate, so it does not matter.
>> also, the soekris can support more than 100 users; i've seen 182
>> unique ip's today, and the day is barely half over.
>> as i'm also a bit of a gps nut, i like the fact that i can run gpsd on
>> my time server to get a view on how the GPS is performing, as well as
>> being able to generate a sky view (http://gpsd.mainframe.cx/). For
>> that matter, I like the fact that I can put perl/php/python on this
>> machine and whip up little scripts as it need them. So a cute little
>> low-power SBC would be neat, but I hope it's not too low-power...
> All that will be possible, you could have access to the NEMA data.
> Is it important?
No way you can maintain that kind of accuracy with "a good crystal",
even an oven controlled one would not do the job. And the average pool
member will not sell his home to buy a cesium osc.
What problem are you trying to solve anyway? We serve time over the
internet, typically with an accuracy of 5 - 50 ms. That is
milliseconds, not nanoseconds, not even microseconds.
We have no need for a 30 ns reference, as it would be a million times as
accurate as the service we can offer to the pool users.
I do believe the pool will need more accuracy in the future, but right
now the network is the bottleneck, not the time reference.
More information about the pool