Re: [time-nuts] Beaglebone NTP server
On 12/11/14, 9:54 AM, Brian Lloyd wrote:
On Thu, Dec 11, 2014 at 10:46 AM, Jim Lux jimlux@earthlink.net wrote:
Actually, I think my point was that the problems I face at JPL are
essentially identical to the problems we face at home. I'm not in the time
and frequency group (and I don't know that they actually are better off..
although they do have rooms full of good clocks), so in our areas, we tend
to have little point solutions to problems.
Has anyone there noticed the amount of time and money spent recreating the
same thing over and over? Sure, individually it is cheap but when you do
the same thing hundreds of times, it is no longer a tiny amount. Clearly it
is a common needed service/device that your IT (or whatever) dept could
provide for nearly nothing. Sheesh.
Nothing in an institutional environment is provided for nearly nothing.
The actual cost to provide a network drop is very close to the
$30/month my project gets charged (there's armies of auditors checking
this stuff).
There's also the phenomenon (applicable to hobbyist and business alike)
of "different kinds of money and resources". As a hobbyist, I pay for
hardware, but my time is (sort of) free, at least in the sense that I
don't have to write a check to myself for the time I spend. I do pay an
opportunity cost in the sense that time I spend stringing coax across my
roof to the antenna is time that I can't spend on something else which
potentially is remunerative.
Similarly, in business, there's a difference between "project funds" and
"infrastructure funds". Likewise, a big difference between "capital"
and "expenses", and local situations and project stuff might push one
one way or another. For NASA, each year stands alone, too, subject to
the vagaries of Congress and the Executive Branch: just because you have
funding this year does not mean that your project will be funded next
year, even if it's a multiyear project.
A 50k project isn't going to be able to buy a $100k Vector Network
Analyzer; they're going to rent one for a few thousand a month. A $10M
project lasting 3 years might say, hey, we'll spend 100k in the first
year, rather than 100k spread over 3 years as monthly rental fees.
Or, more timenuts-ish.. Do I buy a $5k box from MicroSemi, put it in the
rack and go? Do I buy $500 in parts and $1000 in student intern? Do I
spend $3000 in facilities costs to run a fiber drop to my lab from the
maser, the purchase costs of the local cleanup loop and buffers, and
then pay whatever the monthly fee is to keep that fiber alive and my
share of the maser cost? (I actually don't know if the maser cost is
shared by the users, or paid for by an institutional support account..
But for things like network drops, clean rooms, antenna ranges, ESD
inspections, you pay as you use it)
It kind of depends on what my needs are. There are relatively few
things I might do that need the maser, and I might just drag my test
article up the hill to the frequency and timing lab for a few weeks and
pay the labor costs of the FTL folks to do it, and for the rest of the
time when all I need is "reasonably clean quartz oscillator that is
disciplined" I use my local GPSDO (whether XL-GPS or homebrew).
that's why lists like this are so useful.. Homebrew is more common than
some might expect in a research lab environment. And there's an
undeniable benefit in "rolling your own" to develop the skill sets.
I am MUCH better suited to making the make/buy/rent decision because
I've had practical experience getting a GPSDO working, using a truetime
box, and hooking up that cleanup loop from Wenzel to the maser spigot,
and subsequently trying to make ADEV/Phase noise measurements with all
manner of equipment.
So some of that "do it over and over" rather than "institutionally
provided" might be viewed not as paying too much for the ultimate
service, but as a training and educational expense.
This is particularly important as the folks coming out of school are
increasingly specialized (due to the increasing breadth of what might
get covered in a EE or CS degree), and many may not know which end of
the soldering iron gets hot, just because it's not something they ever
had to do.
Which is exactly what we do at work.. Lots of old PCs doing something, but
there's that "what do we do when the PC fails" question. Just like at
home, you've got a limited budget and time available. Do you stock a
couple extra BBBs or RPis? That costs money, and you need to have a place
to store them where you'll be able to find them.
Why not? The freakin' things are down to about $35 for an RPi and $55 for a
BBB. And the price for this is going to go down again. (Well, eventually
fabbing the board and soldering the nearly-free components is going to
define the lower bound in price but the performance can't help but
increase.)
But you also need to stock all the other parts, the GPS receiver, etc.
So you're really looking at something like a few hundred bucks/copy.
And that's a box that has to find a home on a shelf and have someone
remember where it is and what's in it, and to make sure that those spare
parts aren't scavenged for some other project.
And that there is actually the physical room to store them. Every year,
we have to justify the existence of such boxes, because we're all total
pack-rats, and you never know when you might need that old slotted line,
or that piece of unistrut, or that odd shaped piece of waveguide.
While there's not an actual dollar price associated with storage space,
if we were left to our own devices, we'd have no room left in the labs
because the shelves would be overflowing with 40 years of spares.
So we try only to keep things that might get used in, say, the next 10
years, or for which there is a specific use anticipated.
This kind of thing applies to my garage as well as work, except that at
home, I get to be the sole decision maker of "keep vs throw away",
constrained only by the physical space available. If I have that shoebox
with the RPi and GPS receivers in it, do I throw away that S-100 serial
port card to make room for it?
Or do you hope that when it fails 5 years hence, you'll have time and
budget to rebuild?
And in 5 years there will be equivalent or better hardware at the same
price point. Stay in and eat beans for dinner one night to pay for your new
NTP server.
But you'll need to redesign for that equivalent or better hardware. It
is unlikely that it will be a drop in replacement, which is sort of my
point.
You're depending on either a speculative future development/porting
expense, or that the kindness of strangers will provide a cookbook
solution that you can assemble.
Essentially, you want the equivalent of a Microsemi (Symmetricom/TrueTime)
XL-GPS with the right options for, say, $500 or less.
Well, it is looking like I can do this very nicely for about $250 including
the BBB or RPi, LTE-Lite, and a 1U rack-box. Yes I have to build a buffer
for the 1pps and 10MHz output but that has been covered here very nicely.
Yep... and that's way cool.
I do too.
But I'm also not under any illusion that when my homegrown unit fails that
I'll be able to fix it, or rebuild it for the same price. It will depend
on someone having done the redesign using whatever is available then, and
published the cookbook.
I'm still thinking that the spare RPi or two tucked into the box with the
OS already loaded and configured will go a long way toward making sure I
can recover and keep going for many years. I have a BBB with Debian running
NTP that has been running for about 5 months now without a reboot as just a
local NTP stratum-3 server for my local machines. Works pretty well so I am
satisfied that it is stable to do the job. Once it is started I don't
expect to be doing a lot of OS upgrades. If it only runs NTP and all the
other services/daemons are turned off, it is going to be pretty difficult
to exploit. If it ain't broke, don't fix it.
How do you know it's the RPi that will break? Do you also spare the
LTE-Lite? and the buffer amplifier?
So it starts to add up.. An extra $250 today (plus storage space) in
exchange for some speculative redevelopment and $250 in a few years?
My own preference, today, is for the latter... I'd rather learn about
the new technology when the time arises
(My wife, the MBA, would probably say, "put the $250 in some investment
vehicle and in 5 years you'll have enough to pay for the new hardware
AND to pay for someone to hook it up")
Brian...
"Look, I used to leave WWV running all the time on a receiver. That
background DOOOO DOOOO DOOOO was very reassuring as I moved through the
house watching the sweep second hands ticking in time with WWV."
Based on personal experience, this strongly suggests that you weren't married at the time :-).
Tom Holmes, N8ZM
-----Original Message-----
From: time-nuts [mailto:time-nuts-bounces@febo.com] On Behalf Of Brian Lloyd
Sent: Thursday, December 11, 2014 11:21 AM
To: Discussion of precise time and frequency measurement
Subject: Re: [time-nuts] Beaglebone NTP server
On Thu, Dec 11, 2014 at 10:03 AM, Paul tic-toc@bodosom.net wrote:
On Thu, Dec 11, 2014 at 10:27 AM, Dan Drown dan-timenuts@drown.org
wrote:
Is this better or worse than other NTP server platforms? I haven't
tested
them, so I have no idea.
Tharp says his appliance
"can sustain thousands of queries per second. Even under high throughput
timekeeping operations are never disrupted or perturbed."
But enough of that.
From the time-nut perspective what's the interest in high resolution NTP?
I mean beyond the "can I do this?" appeal.
We are all time nuts to one degree or another or we wouldn't be here. I
personally want all the clocks in my house, including all the computers (I
have about 10 running at any given moment here), to have time resolution
better than my ability to perceive errors. (To my eye that is about 100ms
for clocks with a sweep second hand.) and in the 1ms-or-better range for
the computers. I want to know that, when two things get time-stamped on
different machines, I can tell which happened first from the point of view
of a DBMS dealing with concurrency issues.
So a stratum 1 server in my house that is independent of my external
internet connectivity seems desirable.
Look, I used to leave WWV running all the time on a receiver. That
background DOOOO DOOOO DOOOO was very reassuring as I moved through the
house watching the sweep second hands ticking in time with WWV. Time to
move forward in true time-nut fashion.
--
Brian Lloyd
Lloyd Aviation
706 Flightline Drive
Spring Branch, TX 78070
brian@lloyd.aero
+1.210.802-8FLY (1.210.802-8359)
time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.
If there is always a system running, why go with the bbb? I use the file server in the basement fed with 1pps to the non USB serial port from a TBolt in my office.
You do have a file server in the basement don't you? :)
I suppose it would be an interesting experiment to feed an NTP server with several PPS signals (either simultaneous or offset) to see how it handles it. Or to set up an NTP chorus fed from multiple PPS sources or even a single PPS source to see how they average out. Maybe over the holidays. I really should finish up running the coax for a shared GPS antenna to the basement.
Bob
PS: As to the buy a second BBB/RPi question, I've taken to buying what I think will be key spare parts for projects and stocking them. Each project gets a shoebox sized storage bin on the shelves in the basement. One part of the shelves is completed projects, the other is in process. It has the added side benefit of telling me when I have too many unfinished projects. :)
On Dec 11, 2014, at 11:45, "Tom Holmes" tholmes@woh.rr.com wrote:
Brian...
"Look, I used to leave WWV running all the time on a receiver. That
background DOOOO DOOOO DOOOO was very reassuring as I moved through the
house watching the sweep second hands ticking in time with WWV."
Based on personal experience, this strongly suggests that you weren't married at the time :-).
Tom Holmes, N8ZM
-----Original Message-----
From: time-nuts [mailto:time-nuts-bounces@febo.com] On Behalf Of Brian Lloyd
Sent: Thursday, December 11, 2014 11:21 AM
To: Discussion of precise time and frequency measurement
Subject: Re: [time-nuts] Beaglebone NTP server
On Thu, Dec 11, 2014 at 10:03 AM, Paul tic-toc@bodosom.net wrote:
On Thu, Dec 11, 2014 at 10:27 AM, Dan Drown dan-timenuts@drown.org
wrote:
Is this better or worse than other NTP server platforms? I haven't
tested
them, so I have no idea.
Tharp says his appliance
"can sustain thousands of queries per second. Even under high throughput
timekeeping operations are never disrupted or perturbed."
But enough of that.
From the time-nut perspective what's the interest in high resolution NTP?
I mean beyond the "can I do this?" appeal.
We are all time nuts to one degree or another or we wouldn't be here. I
personally want all the clocks in my house, including all the computers (I
have about 10 running at any given moment here), to have time resolution
better than my ability to perceive errors. (To my eye that is about 100ms
for clocks with a sweep second hand.) and in the 1ms-or-better range for
the computers. I want to know that, when two things get time-stamped on
different machines, I can tell which happened first from the point of view
of a DBMS dealing with concurrency issues.
So a stratum 1 server in my house that is independent of my external
internet connectivity seems desirable.
Look, I used to leave WWV running all the time on a receiver. That
background DOOOO DOOOO DOOOO was very reassuring as I moved through the
house watching the sweep second hands ticking in time with WWV. Time to
move forward in true time-nut fashion.
--
Brian Lloyd
Lloyd Aviation
706 Flightline Drive
Spring Branch, TX 78070
brian@lloyd.aero
+1.210.802-8FLY (1.210.802-8359)
time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.
time-nuts mailing list -- time-nuts@febo.com
To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts
and follow the instructions there.
On Thu, Dec 11, 2014 at 1:22 PM, Bob Bownes bownes@gmail.com wrote:
If there is always a system running, why go with the bbb? I use the file
server in the basement fed with 1pps to the non USB serial port from a
TBolt in my office.
You do have a file server in the basement don't you? :)
Yes, I do. Two in fact. They are two of the machines running 7x24 on the
network. Unfortunately I have found that NTP on them does not perform well.
So being able to throw a $55 board at the problem of optimizing NTP
performance and decoupling it from the up/down status of anything else on
the network makes sense to me. I consider routing, DNS, and NTP to be
services of the network infrastructure and should be independent of
anything else.
I suppose it would be an interesting experiment to feed an NTP server with
several PPS signals (either simultaneous or offset) to see how it handles
it. Or to set up an NTP chorus fed from multiple PPS sources or even a
single PPS source to see how they average out. Maybe over the holidays. I
really should finish up running the coax for a shared GPS antenna to the
basement.
That does make sense to me.
Bob
PS: As to the buy a second BBB/RPi question, I've taken to buying what I
think will be key spare parts for projects and stocking them. Each project
gets a shoebox sized storage bin on the shelves in the basement. One part
of the shelves is completed projects, the other is in process. It has the
added side benefit of telling me when I have too many unfinished projects.
:)
That makes sense to me too. As for the 1U time and frequency server, there
is a LOT of extra space in the 1U box for an extra BBB and, potentially,
for another LTE-lite. Of course, the NTP function will work with just a
generic GPS module with 1pps out.
Said: are you going to be doing another run of LTE-lite eval boards? I
think I would like to get one as a spare.
--
Brian Lloyd
Lloyd Aviation
706 Flightline Drive
Spring Branch, TX 78070
brian@lloyd.aero
+1.210.802-8FLY (1.210.802-8359)
On my rpi the jitter is between 2 and 10us iirc but it is a while since
I tested it. A quick peak shows me currently 3us jitter. Not bad for a
userspace solution imho.
Really? The PPS to GPIO interface is handles in user space? It is not
interrupt driven?
Yes it is. But, until today, I used a software solution which
interfaces the interrupt to ntp. I did that because I hate recompiling
kernels and the pps code for rpi+gpio was not in the main distrio.
Folkert van Heusden
--
Ever wonder what is out there? Any alien races? Then please support
the seti@home project: setiathome.ssl.berkeley.edu
Phone: +31-6-41278122, PGP-key: 1F28D8AE, www.vanheusden.com