[sdnog] a question from an sdnog attendee

[Nishal Goburdhan]

On 18 Aug 2016, at 11:28, <someone? wrote:

> you told us about the raspberry pi stratum clock, and it's possible to
> deploy 10 RPi clocks around your home country. How reliable is this? 
> how
> realistic it is to use such a simple method to acquire stratum 
> accuracy for
> a serious network?
> thanks, and regard :)

[i received this question in private, but i’m replying to the mailing 
list since it would be a good answer to all]

let’s start with the base assumption;  as i explained, this is *not* a 
reliable clock - this is a cheap clock.
but, we can achieve a better degree of reliability via deploying 
multiple of these.

if you can find places such that:
* unrelated power
* different IP networks/transit providers
* perhaps different operating systems (philip would tell you to run 
freebsd, instead of raspian  :-))
.. and build this, such that there’s no real single point of failure 
that you control, i believe it’s possible to get serious enough 
accuracy.

of course - using GPS, always puts you in the risk factor that you’re 
using someone else’s satellites :-)

recall that the rPI itself, probably isn’t what you want users to 
synchronise to;  you want them to pull time off a faster, more powerful 
stratum-2 server, which is synced to your S1.  and, in this manner, you 
can protect the “weak link” in the rPI (the poor CPU and network 
performance).

would i do this for production?
yes!   i have three (3) stratum-2 clocks that we make available for 
general public use:  ntp[1-3].inx.net.za.
these are fed from some domestic, and some international sources.  i 
plan to deploy a bunch of my own S1 clocks, and have my S2 clocks, sync 
to my *cheap* S1s, and other domestic S1s too ..

—n.