[time-nuts] SLIP vs Ethernet for NTP

[Jim Lux]

On 10/23/11 6:15 AM, Poul-Henning Kamp wrote:
> In message<4EA411F5.1030303 at rubidium.dyndns.org>, Magnus Danielson writes:
>> On 10/23/2011 02:37 PM, Tijd Dingen wrote:
>
>> The danger of Ethernet is that it has high capacity and a interconnect
>> friendly interface. Thus, you might feel inclined to toss data over it
>> carlessly causing packet delays and you can hook it into a switch and
>> get delays and packet losses there.
>
> I usually cure that by using fiber-based ethernet, which leaves a lot
> less to plug it into.  Also firewall rules preventing anything but
> port 123 packets helps a lot.
>
> RS-232 has surprisingly bad jitter.
>

if you're talking asynchronous RS232 (the by far most common, these days)
off hand, I'd expect the jitter to be on the order of 1/8 bit time, 
uniformly distributed.  An awful lot of UART implementations generate a 
8x clock to sample the input and find the rising edge of the start bit.


But if you had an old synchronous RS232 you could use Rx Clock (on pin 
17 of the 25 pin connector, having just looked it up).  Back in the 
70s/80s when these things were still common, the timing was pretty 
stable (since you didn't have a lot of fancy hardware to adjust things 
and buffer memory was expensive)

There would still be issues with the slew rate, etc.


And if you have issues with Ethernet, then IEEE-1588/PTP might be for 
you.  Time tagging the sync pattern of the ethernet frame with an 
onboard oscillator so you can get rid of any latency effects in the 
software stack.