Todd wrote: > Hal Murray wrote: > >>> I have started to see this 5mhz frequency quartz oscilators that are >>> really inexpensive (see http://www.alltronics.com/cgi-bin/ >>> category.cgi?item=04P010 ). My question, how would you get this to >>> work with ntpd? Seems like it work great for my situation where I can >>> just put inside the rack behind the server, use some ntp servers to >>> get the time to within 1 msec usually and then use this to get into >>> the nanosecond accuracy. >>> >> There are two issues with keeping time. >> >> The first is to get your clock to tick at the right frequency, or to know >> how >> fast your clock is actually ticking. >> >> ntpd is pretty good at that. If you haven't done it already, turn on >> loopstats and graph the offset and drift columns. The drift is basically >> the >> difference between the nominal frequency and the measured frequency, aka the >> inaccuracy of the crystal. Most crystals make reasonable thermometers. >> >> The other issue is lining up the seconds ticks. You can get reasonable >> results by asking nearby ntp servers that know the answer. How good you get >> depends mostly on your network connection. If you want better than a few >> milliseconds, you need a local PPS source. >> >> > > So I pretty much had things backwards? using the PPS for the "tick" and > then network ntp servers to get it pretty close to the real time. > > >> If you want to get started, I recommend the Garmin GPS 18 LVC. Under $100. >> Some soldering required. Not much. (It needs 5V. You can get it from USB.) >> http://support.ntp.org/bin/view/Support/ConfiguringGarminRefclocks >> I got mine at ProVantage.com There are some on eBay, but the picture is the >> USB version. >> >> You want the OEM version - without maps or software. >> > > Yeah I've looked into that. My problem is the server I want to do the > really precise measurement of is located in a secure datacenter that > they're not going to let me drill a hole in the roof just so I can get > really really accurate time when really ntp works just fine. I may end > up looking into it at home, but even if I run it at home and then use it > as a source for the server at the datacenter you then introduce the > network latency into it and I pretty much lose any benefit of using a > local time source. That's why I saw this little crystal oscilator and > thought that could be used for a PPS signal. > > For the curious: http://rikku.vrillusions.com/ntp/ (i have a bunch of > "noselect" ones I'm just evaluating network latency and such) So > basically just from ntp servers it stays under 1msec usually although it > has been a little more bumpy recently. > > What about making use of cell phone network timing? Can you actually receive/use a cell phone within the data center?
Bruce _______________________________________________ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.