[manchesterwireless] Extreme solutions...

[Harry Moyes]

On Tue, 2005-08-02 at 20:52 +0100, G.W. Haywood wrote:


> 
> Hmmm...  250/186411 = 0.00134 ... How do you think he gets his signal
> to go faster than light?  :)


I like the colourfully explanations better, but if you want a guess
at the real cause try the following.

 
Sadly this is a common instrumentation error. The timer resolution in
the OS is limited, and there are two passes at least through the
scheduling algorithm.



submit ICMP frame to radio    >    >    >          Frame reaches target
  submit time stamp request to OS
    OS reschedules
       timestamp reply  ^

response frame received        <      <     <           Target Responds
  submit time stamp request to OS
    OS reschedules
              ho hum doing something else
                                         timestamp reply  ^


                       ^  error in reported transit time  ^



Or the other way round........



submit ICMP frame to radio    >    >    >          Frame reaches target
  submit time stamp request to OS
    ho hum doing something else
                                    timestamp reply  ^


response frame received        <      <     <           Target Responds
  submit time stamp request to OS
    OS reschedules
        timestamp reply  ^


                         ^                           ^
                      negative error in reported transit time



At a resolution of tens of microseconds present standard Linux Kernels
are struggling. Linux kernels usually report ping times in tenths
of milliseconds.

Most versions of windows only report ping times in whole milliseconds.

Now with Ingo Molnars shiny new RT Linux patchset maybe you can get
down to this sort of interval reliably..... 


Harry