[ntp:questions] All my servers rejected
David Moerman (dot at dot)
david.m at advmeas.com
Wed Jan 7 20:14:00 UTC 2004
I've got a situation where I'm trying to synchronize through a modem
connection (115200 bps through a satellite). Ping delays are between 2 to
3.5 seconds, and jitter is often above 1 second. I've tried between 3 and 5
different servers; I've tried using just NIST servers; I've tried just
stratum 2 servers with varying NTP version (3 or 4) support; no difference.
BTW, I'm using ntpq version 4.1.2 on linux 2.4 kernel (redhat 9).
Actually, I *have* had success *once*: it took between 60 and 90 minutes to
synchronize (not sure how long it took because I had given up and forgot
about it, then checked again, surprise!). Also, I checked my bandwidth
usage, and it was in the hundreds of kilobytes. This is definitely not
acceptable. ntpq reports that all the servers are connected, but the
_space_ tally code shows "reject".
My situation is this: my modem connection is usually off (could be for an
hour or more), then connects for about a minute (PPP connection) and then
goes off again. I would like my time sync to happen within that period, and
sync regardless of my offset (I have verified it is using ntpd -g). Also, I
don't want high bandwidth usage, because that drives up my satellite
air-time costs. I tried using ntpdate, and it works fine, and used only
about 1KB total. ntpd version 4 claims to be able to duplicate this with
the 'iburst' flag, but this does not seem to work (do all the servers need
to be NTPv4?). Why shouldn't I just use ntpdate? That would be fine, but I
also have a client sub-net connected which will be using my ntpd as a
server, so it will be running anyway. I'd like to get it to work and not
use ntpdate if possible.
One last thing: this is an automated, headless system. It must run with no
administration for long periods. I absolutely require reliability.
Advanced Measurements Inc
More information about the questions