ntpd: program structure

[Achim Gratz]

Hal Murray via devel writes:
> The current code is a combined client and server.  I think we should split 
> that into separate programs.  Maybe not right away, but we should at least 
> start thinking that way.

If you're going to give ntpd the djb treatment (ref. qmail) then that
split is the most obvious one, but I suspect you shouldn't stop there.

> The server side is simple.  It gets time from the system.  There are a few 
> other parameters that either need to come from the system (via ntp_adjtime?) 
> or from the server via shared memory or similar.

That's been done already as a POC, written in Rust:
https://github.com/mlichvar/rsntp

> We should be able to run multi-threaded with only minimal changes.  It would 
> need locks for updating various statistical counters.  There are probably 
> others, but I can't think of them.
>
> We would probably want to restructure things a bit.  For example, a thread (or 
> several) would be associated with a socket rather than going through a layer.

It seems that as with all multi-client IO-bound and latency sensitive
code you should employ just enough CPU cores to be able to saturate the
interface.  Actually I think that most of the NTP server on a Linux box
should be done by the kernel, optimally via XDP / eBPF programs.

> The client is pretty much what we have.  Just ignore the small amount of 
> server code.  If we run it on another port, we can run it at the same time as 
> a server.

There is most of what a client needs already in Go:
https://github.com/google/gopacket/blob/master/layers/ntp.go

A client not using this layer is here:
https://github.com/beevik/ntp

> I'm not sure what to do with ntpq.  Most of what I use it for (peers, -p) is 
> talking to the client but the server would be listening on port 123.
>
> Anyway, I think that thinking about them as separate parts will help our 
> discussions.
> We should be able to improve performance on busy servers.

It's been decades since I looked at an NTP server that has enough
clients to make me wonder about performance, so I'd like to see actual
numbers for a busy NTP pool server w/ a fast connection for a baseline.

Here's some code to test with synthetic load:
https://github.com/mlichvar/ntpperf


Regards,
Achim.
-- 
+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

SD adaptations for KORG EX-800 and Poly-800MkII V0.9:
http://Synth.Stromeko.net/Downloads.html#KorgSDada