[time-nuts] Advice on NTP server needed

[Chris Albertson]

OK so the USB works at the ms level.   That compares to the us level
I'm getting.  So I have 500 to 1,000 times better performance using an
$85 Intel Atom board.  That's $85 with the CPU and the serial port
soldered down on-board.  I did need to add a 1GB RAM and a micro-atx
size case but I'm still under $200 for the entire server.   I was able
to get it all from Amazon.com and not have to hassle with eBay

If I had to spend a ton of money to get that 500X better performance
then I'd think about it I did not pay extra

I can get ms level performance using pool servers on the Internet,  If
milliseconds is all you need then don't bother with GPS. but then I
have an Ethernet equivalent Internet connection.   You should expect a
handful of us error from a GPS driven NPT server and ms level  from
Ethernet connected clients

If you are willing to invest much effort you can do better than 1uS.
THAT would be in the Time Nut Range

On Mon, Jun 13, 2011 at 3:59 PM, Attila Kinali <attila at kinali.ch> wrote:
> On Mon, 13 Jun 2011 15:37:13 -0700
> Chris Albertson <albertson.chris at gmail.com> wrote:
>
>> For NTP the issue is different.   The PPS signal coming from the GPS
>> needs to go into a hardware serial port so the PPS handler sees a very
>> low but more importantly a predictable latency.    The PPS depends on
>> a very specif interface to the hardware interrupt controller.
>
> The interrupt latency for USB is predictable. You can read out the
> bInterval setting of the interrupt endpoint of your serial converter
> (use something like lsusb or similar), this is the maximum poll interval
> in ms (for USB FS devices). Most devices i've seen, set this to 1
>
> Given that USB port is fully loaded, then you will get an average
> delay of 0.5ms (uniform distribution). With an unloaded USB port,
> this will be much better. (for exact values, check the source of
> your OS/driver or use an USB debugger like the beagle to measure it)
>
> NTP will filter out the jitters of the PPS input using its control
> loop, so in the end, all that's left is the average delay as offset.
>
> Considering that anything below 1ms is negligible when comming to NTP,
> then i'd say that a USB serial port is good enough. Of course, 1ms timing
> uncertainty is not time-nuts class, but then again, ntp isnt
> time-nuts class ;-)
>
>                        Attila Kinali
> --
> Why does it take years to find the answers to
> the questions one should have asked long ago?
>
> _______________________________________________
> time-nuts mailing list -- time-nuts at febo.com
> To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
> and follow the instructions there.
>



-- 

Chris Albertson
Redondo Beach, California