[time-nuts] Raspberry Pi tweaks and custom kernel, was RE: PPS for NTP Server - How Close Is "Good Enough"?
m.matthew.george at gmail.com
Sun Jun 14 00:26:41 EDT 2015
Ed, I finally found some time to read David's post more closely. David's
documentation on his website has been invaluable in my interest with NTP
and the raspberry PI. The detailed monitoring he has done and the captured
data has been so helpful as I have been working with the Raspberry PI 2 and
NTP. Spend some quality time on Daivd's pages... good stuff and the
number speak volumes.
I must agree with David's last comment related to the latest Raspbian
kernels. On the PI 2 the compile time is much better, however even with my
tweaks, I don't have the documented data to back up my observations. Which
makes my observations antidotal. Lean on David's data as a base line if
you want to duplicate his performance.
I'm going to put together a more detailed description of all the changes I
have made and post it to the list here or provide a link. For me, this has
been an adventure in getting NTP to show nano second level consistent
offsets and learning as I go. Providing another NTP server to the NTP pool
is fun too and it's fun to see how much traffic the PI can easily handle.
I totally get it that users on the other end will never get and or care
about my sub microsecond offset obsession... it's just fun to chase it...
nothing more. It is fun however to run iptraf and see the flood of UDP
traffic my PI 2 can handle as it serves up NTP time sync requests. And at
the same time run top and see that ntpd is using 1% of the cpu as I mention
On the temperature front, with www.pool.ntp.org bandwidth set at 25MB (just
a relative load indicator based on the docs at pool.ntp.org), ntpd shows
that it's using about 1% of the cpu on average. As for the temperature of
the CPU, it's very low in the grand scheme of things at 44c at this trivial
pi at raspi2 ~/dev/frequencytest $ /opt/vc/bin/vcgencmd measure_temp
I have a tiny passive copper heatsink on the cpu, but at this load, I don't
think it drops the cpu temp much. It's all running headless with nothing
connected to the USB ports. I have no need for it for this application.
Anyway, I'll try to get motivated to do a full writeup... I can't touch the
work David has done however! He is being way too modest about his work and
On Sat, Jun 13, 2015 at 12:06 AM, David J Taylor <
david-taylor at blueyonder.co.uk> wrote:
> Hi Max! Thanks for the information, I was wondering if you had documented
> what you did to your Raspberry Pi so that it might be reproducible to
> someone like me (a newcomer time-nut and intermediate Linux user) ... you
> had said:
> Thanks so much for your assistance! Sorry if these questions have been
> posted before, but I am very curious about your setup as it nearly matches
> -Randal "r3" of CubeCentral
> I would also like to know how to best tweak the Raspberry Pi 2 for best
> performance as an NTP server. Although I have been using Linux-based
> firmware in my routers for several years now, I have never actually
> worked In Linux before.
> I have gone through several tutorials on compiling a custom kernel, only
> one or two have actually ended in a compilation, but then I couldn't
> figure out where the kernel and modules were and get them onto the Pi. I
> have tried this under Ubuntu x64, Debian x64, and Mint 32 bit. So far
> Mint has been the best, I successfully compiled and I found the kernel,
> but cannot figure out where the modules are. Anybody suggest a really
> nice tutorial for learning this Linux stuff? My experience so far is
> really leading me to appreciate Windows.
> Randal, Ed,
> Just in case you missed it, I am also a beginner to Linux, so I documented
> my steps to get NTP working on the Raspberry Pi on my Web site. There's a
> quick-start guide here:
> and a more blog-like set of detailed notes here:
> I hope they may be of some help. Typically I'm seeing well under 5
> microsecond offsets reported by the RPi themselves:
> although with different GPS devices, antennas, and system loads the
> performance does vary. The polled USB is, perhaps, the main limitation to
> the device as a server on the LAN, although my own tests have shown RMS
> offsets reported by a remote client on a quiet LAN of 39 microseconds and a
> jitter of 38 microseconds. Likely on a busier LAN the network itself might
> be the limiting factor.
> I am no longer convinced that there is a significant gain to to be had by
> recompiling the kernel, now that PPS support for interrupts is included in
> the current Raspbian kernels. I'm willing to be convinced otherwise,
> though. Recompiling is a long and painful process, and cross-compiling
> presents further problems!
> SatSignal Software - Quality software written to your requirements
> Web: http://www.satsignal.eu
> Email: david-taylor at blueyonder.co.uk
> time-nuts mailing list -- time-nuts at febo.com
> To unsubscribe, go to
> and follow the instructions there.
More information about the time-nuts