[ntp:questions] latencies if going through Linux parallel or serial port drivers?
Ulrich.Windl at RZ.Uni-Regensburg.DE
Thu Apr 30 12:38:05 UTC 2009
Unruh <unruh-spam at physics.ubc.ca> writes:
> I am wondering if anyone has investigated what kind and magnitude of
> latencies one gets by going through the parallel and serial port drivers on
> Linux? I have a module, gps, which attaches to the parallel port interrupts
> to deliver time to one of my systems. I seem to be getting latencies around
> the 1usec region (bot from teh jitter on the timing and from the direct
> measurement of the time between asserting pin 9 on the port and the receipt
> of the interupt-- ie use do_gettimofday just before writing to the parallel
> port register and getting the interrupt.
> I have not tried using the standard driver modules on the system
> (parport_pc or serial modules) to see what kind of latency they introduce
> and was wondering if anyone else has.
> One reason for asking is that if the parport_pc is loaded, when I unload it
> to install my module, that unloading leaves the parallel port switched off
> and useless, and a reboot is in order. Now there may be a way of switching
> on that port again, but I have not been able to delve into the intricacies
> of pci programming enough to know how.
I can only tell you that I've implemented the "echo" feature of the PPS
API in my PPSkit for Linux, and for the (now sleeping) 2.6 version I
think (I vaguely remember) to have implemented the feature for the
paralel port as well. If you cut the echo response time into two halves,
you get an estimate of the delay and jitter. Doesn't even my FAQ have a
sample graph? ... memory...
More information about the questions