[rescue] In Realtime: Saving 25, 000 Manuals ??? August 15, 2015
rdarlington at gmail.com
Fri Aug 21 12:11:35 CDT 2015
I'm going to make boards available, gps modules, eagle files, etc. But in
full disclosure, I got married 2 months ago and haven't had any time to
work on it since then.
What I see with the BeagleBoneBlack (BBB from here out) is that it is
sensitive to CPU load which translates into temperature issues. I was
going to heatsink the CPU and run some tests. It works wonderfully just
sitting there being an NTP server but as soon as I crunch the data and
create graphs it spikes the load. I wait 5 minutes and I can see a spike
in the offset when creating new graphs.
I suspect if I could keep the load up I can steady this out. NTP will
happily adjust. It's the transients that effect it. NTP servers really
like to be in constant temperature vaults but that's not practical. Next
best thing is a cardboard box with a fan blowing on it as needed. Or
styrofoam beer cooler packed with water bottles. A combination of the two
maybe. Things like that.
Once I got NTP working such that I was mostly happy with it, I moved onto
getting a drive bay sized graphics LCD screen working. I wanted to plot
stats with something like MRTG:
So yeah, another $60 but there is certainly no requirement to use something
like this. I really was just hell bent on replacing my Symmetricom unit
with something far better for far less money and this was still in the
Making this thing work at all is a royal pain in the butt. I did have it
working but got sidetracked and after coming back to it, it seems I can't
reproduce it on this little BBB computer. Worked fine from a linux VM
though so I know it will work given enough effort.
In regards to jitter on the usb and serial lines, the time-nuts crowd would
scoff at this. We're used to dealing with time scales in the tens of
nanoseconds to much, much lower. Jitter on the serial lines is in the
milliseconds. Does it matter for NTP? Probably not.
To get around any issue with jitter on serial, be it usb or rs232, you use
the PPS line. But how? Well, gpio lines right into a CPU works about as
good as theoretically possible. Nanosecond jitter levels. PPS going into
a serial port brings you right back to millisecond to hundreds of
microseconds and that's starting with something with tens of nanoseconds of
jitter. Basically taking something useful and making it 4-6 orders of
But what's useful? Do our filesystems use anything better than one second
precision? Surely a few hundred microseconds won't matter. And it doesn't
unless you're a time-nut. These guys are used to measuring distance in
nanoseconds (in particular, wire and cable lengths). How temperature
effects measurements. It's very common for me to notice significant
differences in measurements solely because of a temperature change in my
lab. It's a pain.
I recommend staying away from that list as it's a slippery slope into
insanity pretty quickly. For years now I haven't been able to look at a
measurement standard without wondering "says who"? and "how do you know
On Fri, Aug 21, 2015 at 10:49 AM, Lionel Peterson <lionel4287 at gmail.com>
> I, and I assume others would be interested in having our own time source
> me the issue purely for geek points)... I'm interested in anything that
> can be
> cheaply cobbled together to accomplish that end.
> In related news, there seems very little support for the actual NTP
> project -
> been reading apocalyptic stories about it's demise these last few years...
> Seems NTP isn't 'sexy enough' to garner any real support, though many
> rely, either directly or indirectly on accurate time.
> > On Aug 21, 2015, at 12:04 PM, Bob Darlington <rdarlington at gmail.com>
> > Yes, we do have discussions on that but ultimately that's probably a
> > project for the ntp oriented groups.
> > I'm designing a cape for a beaglebone black around a Furuno gps timing
> > receiver in order to replace a commercially available ($3000) NTP
> > appliance. So far I'm out about $100 including the beaglebone. The gps
> > receiver, brand new, is $35. Ping me off list and I'll give you all the
> > details you want.
> > Here's some random stats for a 24 hour period that I bothered to graph:
> > http://www.nerdhouse.org/ntp/
> > time-nuts is much more interested in the science and art of
> > performance of oscillators, precision time interval measurements,
> > statistics, and the like. Most of us care about what time it is
> > (wristwatch time) but that's a minor interest. It's all about squeaking
> > out the best performance of cesium beams, rubidum clocks, and quartz
> > oscillators for stability.
> > -Bob
> rescue list - http://www.sunhelp.org/mailman/listinfo/rescue
More information about the rescue