[LinuxPPS] time compensation

[Udo van den Heuvel]

On 2010-06-02 16:42, Hal V. Engel wrote:
> On Wednesday 02 June 2010 07:35:33 am Udo van den Heuvel wrote:
>> On 2010-06-02 16:12, Remco dB wrote:
>>> Milliseconds of delay therefore are not the result of coaxial cabling in
>>> my opinion. How the receivers process the GPS signals and 'convert' them
>>> to your DCD-pin, is of course another issue.
>>
>> As the Garmin GPS18LVC does only have a serial connection to the PC,
>> this part of the trajectory is most of interest w.r.t. delays.
>> A conservative estimate could help here.
> 
> About 5 nanoseconds per meter is about right.

OK.
I have about 5 meters extra cabling plus 3 from the GPS18 itself.
Then the RS232 port delay and we're in the kernel.

The ntp refclock doc says:

time2 time
    Specifies the serial end of line time offset calibration factor, in
seconds and fraction, with default 0.0.

So I set for `time2 0.00000004` ?
(to compensate for just cabling)
Or a negative value?

Then what is the irq latency?
Can I measure it someway?
Should I run LinuxRT on this simple single CPU Epia?

Kind regards,
Udo