Just ran a test in which I turned off chrony and ntpd (not running) and monitored the offset of a Sure GPS connected to the ACK pin of my parallel port, and a parallel interrupt service module I wrote. I got brief blips in which the offset was clarly 10-40us late. Ie, something is eating up interrupt time-- delaying the interrupt by 10-40us. and these occurred at clustered times. The usual fuzz in the offset was in the few us range, and those popcorn blips were pretty rare. (maybe 20 in half a day-- interrupts were once per second. ) They were also isolated (ie, you would not have two blips side by side. ) The median filters in at least some of the refclock drivers would have removed those blips. And with a poll of 4, (ie, about 10 samples after median filtering), the level of the offset noise per poll would be at the 1us level.
_______________________________________________ questions mailing list [email protected] http://lists.ntp.org/listinfo/questions
