[189275] in North American Network Operators' Group
Re: NIST NTP servers
daemon@ATHENA.MIT.EDU (Mel Beckman)
Fri May 13 16:38:47 2016
X-Original-To: nanog@nanog.org
From: Mel Beckman <mel@beckman.org>
To: Lamar Owen <lowen@pari.edu>
Date: Fri, 13 May 2016 20:38:41 +0000
In-Reply-To: <57360102.9060809@pari.edu>
Cc: NANOG <nanog@nanog.org>
Errors-To: nanog-bounces@nanog.org
Lamar,
Because you need microsecond-level time accuracy (which is beyond NTP's cap=
abilities) you'll requires an adjunct protocol, such as PPS, to get that. =
For continued NTP delivery despite periodic GPS signal loss, then you need =
an OCXO internal clock.=20
But anyone satisfied with NTP's millisecond time accuracy at worst needs a =
$1 temperature-compensated internal clock. Either method needs the specs fo=
r a Stratum 1 time source on a local network.=20
As you point out, the hobbyist SBCs can't deliver even basic clock accuracy=
. =20
But another key consideration beyond accuracy is the reliability of a serve=
r's GPS constellation view. If you can lose GPS sync for an hour or more (n=
ot uncommon in terrain-locked locations), the NTP time will go free-running=
and could drift quite a bit. You need an OCXO to minimize that drift to ac=
ceptable levels.=20
But I see that the price premium for an OCXO clock is only $100 to $200 on =
low-cost (I.e., ~$500) commercial NTP servers. So buyers need only make a m=
inor cost adjustment to have very good, and inexpensive, COTS NTP performan=
ce and reliability.=20
-mel beckman
> On May 13, 2016, at 9:30 AM, Lamar Owen <lowen@pari.edu> wrote:
>=20
>> On 05/13/2016 10:38 AM, Mel Beckman wrote:
>> You make it sound like TXCOs are rare, but they're actually quite common=
in most single board computers. True, you're probably not gonna find them =
in the $35 cellular-based SBCs, but since these temperature compensated osc=
illators cost less than a dollar each in quantity, they're quite common in =
most industrial species for well under $100.
>=20
> Correct, they're not rare in the industrial line (for that matter you can=
get TCXO-equipped RTL-SDR dongles, but that's not NTP-related). Something=
like a Transko TFC or TX-P or similar is enough for reasonable timing for =
basic purposes, and they're not expensive. They're also not a stock item o=
n the consumer-level SBC's either. I looked at one of our half-dozen ODroi=
d C2's, and the main processor clock, X3, is under the heatsink, so I can't=
see what part is being used. X1 and X2 are outside, and it doesn't appear=
that they are TCXO modules, although I didn't use a magnifier to check the=
part number and might have made an error.
>=20
> The Nicegear DS3231 RTC has a TCXO, and might be the best low-cost choice=
at $12 (need to have an RPi, ODroid, or similar on which to mount it). It=
's not that TCXO's are rare or expensive, it's that they're not often consi=
dered to be important to accuracy in many circles.
>=20
>> An Ovenized XCO is absolutely not required for IT-grade NTP servers.
>=20
> No, but it is for my purposes here. But, as I said in my post:
>=20
>=20
>> You really have to have at least a temperature compensated quartz crysta=
l oscillator (TCXO) to even begin to think about an NTP server, for anythin=
g but the most rudimentary of timing. Ovenized quartz oscillators (OCXO) a=
nd rubidium standards are the next step up, ...
>=20
> I was just saying that OCXO and Rb are just the next step up if you would=
like more stability, that's all. For 'within a second' on a GPS-disciplin=
ed clock (even without the 1PPS signal) you wouldn't necessarily need TXCO.=
But that's what I meant by 'anything but the most rudimentary of timing.'=
Rudimentary is down to the millisecond in my environment. PTP takes you =
to the next level, and beyond that you don't use network timing but put a d=
edicated time distribution network running IRIG-B or similar. But that is =
beyond the scope of a typical IT NTP server's needs.....
>=20
