[ntp:hackers] ntp and signalled IO

Paul Vixie paul at vix.com
Tue Apr 18 17:01:21 UTC 2006


> >So, is there a way in eventlib to prioritize the events in this
> >manner?
> 
> Not that I know of.

i'm not convinced that we can predict or subtract-out the activation time
of a physical i/o event if select() or poll() is used, with or without
prioritization, with or without an eventlib shim above the system calls.
for that matter i don't see a way to predict or subtract-out SIGIO latency.

when i did my most recent refclock_goes work, i put a counter-timer board
in the computer (grep for "#ifdef PCL720") and triggered it with the on-time
TTL so that i could figure out, once the SIGIO finally activated my read(),
how long it had been since the ascii mantissa i was reading had been correct.

if SO_TIMESTAMP means the kernel can put a timestamp on an i/o event, and
it includes serial i/o, then i'm all for it.

if we can generalize my PCL720 hack (i know that some folks use the line
printer parallel port for the same general purpose), then i'm all for THAT.

if we can do it all with PPS_API then i'm all for THAT.

but eventlib won't make any timings worse, it'll just make the code better.


More information about the hackers mailing list