Paul Theodoropoulos paul at anastrophe.com
Mon Mar 11 03:52:02 UTC 2019

On 3/10/2019 6:45 AM, NTPfiend wrote:
> OK, replies noted carefully.
> CPU governor, new territory for me.   R Pi documentation says the 
> default is "ondemand":
> https://www.raspberrypi.org/documentation/configuration/config-txt/overclocking.md
> which at first sight is a problem.   But, my R Pi has the upper and 
> lower limits both at 700 MHz, so unless the mere task of ramping nowhere 
> takes cpu oomph, there should be no difference from "powersave" or 
> "performance".   I could edit /etc/init.d/cpufrequtils or possibly 
> /etc/init.d/raspi-config (not the usual raspi-config) perhaps.

You can set boot cpufrequtils default in /etc/default/cpufrequtils. Back 
in my early testing I just dumped all the possible schedules into the file 
so it would simplify changing -


But that would all appear to be moot, as you're already on performance, 
and a more limited older chip. So the solution really would be offloading 
(but note caveat below...)

> To use ntpviz on a remote machine, I'd have to transfer the files. It 
> appears simple (to me) to rsync from the ntpsec R Pi to somewhere else, 
> with rsync being nice'd, maybe ionice'd and well throttled back with the 
> --bwlimit option.

I use the rsync method to copy logdata, and I've never found a correlative 
spike in just doing that data xfer. It's chugging through it all with 
ntpviz that drastically heats up the board - which heats up the crystal - 
poof - spikes.

One way to soften that - without offloading - would be to enclose the RPI 
in an insulated box, and run the ntpheat utility (thanks Gary!) to try to 
keep the board temp at roughly the same temp it would be in-between runs 
of ntpviz. You'll still get variations across days and weeks as ambient 
temps and internal temps try to find equilibrium, but that's unavoidable, 
and still more desireable than hard spikes.

See this series of blog posts (thanks again, Gary!) for vast amounts of 
insight on all matters heat, raspi, ntpsec.


Paul Theodoropoulos

More information about the users mailing list