[ntp:questions] ldattach makes ttySx buffered !?

unruh unruh at wormhole.physics.ubc.ca
Sun May 2 15:46:50 UTC 2010


On 2010-05-02, Richard B. Gilbert <rgilbert88 at comcast.net> wrote:
> Wolfgang Breyha wrote:
>> Hi!
>> 
>> I'm running a soekris net4801 with openwrt 10.03. kernel 2.6.32.10 with
>> LinuxPPS patch and ntpd 4.2.6p1.
>> 
>> Today I recognized that my /dev/ttyS1 gets a input buffer as soon as I run
>> ldattach PPS /dev/ttyS1
>> 
>> The buffer is hold until somebody connects to /dev/ttyS1 and reads it. This
>> is reproduceable with "cat" at any time.
>> 
>> ntpd gets confused by the input at start big time and needs really long to
>> settle.
>> 
>> I have not found the cause of this yet. Has somebody noticed similar behaviour?
>> 
>> Greetings, Wolfgang
>
> NTPD has always needed a long time to settle.  From a cold start you 
> need to wait ten to twelve hours for the time to reach the best possible 
> accuracy!  This is not a great problem if you keep NTPD running 24x7.

Actually it is a problem for the best clock discipline, 
 because it means that ntp also reacts very slowly to
changes like temperature chages.

>
> For those who need/want to turn NTPD on at 0830 local time and turn it 
> off at 1700 local it's a real problem to which the only solutions are: 
> to suffer, to run 24x7, or to use a tool better suited to the circumstances.

Like running chrony if you are on Linux or bsd. There is nothing in the
fundamental structure of ntp which causes the attrocious settling
behaviour, but certain design decisions in using the ntp time gathering
to discipline the local clock. Mills regards his stability analysis as
more important than reaction time.
However it is not clear that this is what is causing the OP's
problems.It may be, or there may be aproblem with the serial port
buffering.

>




More information about the questions mailing list