[email protected] said: > Yes, although from some GPS devices the jitter may be worse than from > Internet servers (depending on your connection).
I've been looking for "good", low cost GPS gizmos, preferably with no soldering required. If anybody finds one, please let me/us know. The best I've found is the Garmin GPS-18x. It is only "good" if you can use the PPS signal and it requires some soldering. All the low cost GPS unit's I've tested have horrible jitter. They are crappy without PPS support. In particular, the USB units don't have anything like PPS. I'd call them good-enough if they worked as well as I hope. The USB jitter is not a problem, at least for some/many people. It's small relative to network delays. I'd consider a USB device good enough to be interesting if it worked as expected. The problem isn't just jitter, it's wander. By that I mean very low frequency that's hard to filter out. Ballpark number is 25 ms of jitter and 100 ms of wander. These are from SiRF chips on USB: http://www.megapathdsl.net/~hmurray/bb/gps/Holux-2.png http://www.megapathdsl.net/~hmurray/bb/gps/BU-353-gpgga.png The older Garmin GPS-18 (non x) was good enough. (but, unfortunately, not as sensitive and no longer available) http://www.megapathdsl.net/~hmurray/ntp/GPS18USB-off.gif On the network.... I have a 384K DSL line. It's mostly idle. ntpd has no trouble filtering out the occasional poor samples when it happens to collect data while I'm loading a big web page. On the other hand, I occasionally download CDs or such. That takes hours. That ties up the line for long enough so that the queuing delays can confuse ntpd. I've seen delays of 3.5 seconds. There is a bufferbloat project working on that area, but it's going to be a lot of work. http://www.bufferbloat.net/ It's screwing things like VoIP. If/when it gets fixed (or even improved) timkeeping will get better for free. -- These are my opinions, not necessarily my employer's. I hate spam. _______________________________________________ time-nuts mailing list -- [email protected] To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.
